§ 01 · The Hook
The bait, then the rug-pull.
By the time most business owners realize their AI stack is broken, they have already paid for it twice — once in software subscriptions and once in the hours their team spends being the human bridge between tools that were never wired together. This playbook starts where most tutorials skip: not with a cool demo, but with a diagnosis.
Where the time goes.
00:00 – 02:02
01 · Cold open and client proof
Promise + $3M client case study: lead response 47h to 60s, close rate +35%, 12h/week data entry eliminated.
02:02 – 05:19
02 · Why most businesses use AI backwards
Frankenstein stacks, 42-tool average, data silos, team as human bridge. McKinsey: 88% adopt AI, only 1% make it work across the business.
05:19 – 08:59
03 · The 5 Pillars of an AI-First Business
Clean data, intelligent workflows, connected systems, agentic operations, real-time visibility. Miss one, it wobbles.
05:19 – 12:32
04 · The 3 Eras framework
Era 1 (manual), Era 2 (siloed tools — where most businesses are), Era 3 (agentic infrastructure — the destination).
08:59 – 16:19
05 · Operator-to-owner shift
Era 3 = 30-minute morning strategic review instead of 8-hour catch-up. Systems run without you.
16:19 – 20:07
06 · Why 80% of AI projects fail
Rand: 80% never reach production. MIT: 95% of generative AI pilots show no measurable bottom-line impact. Root causes: doing too much at once, wrong sequence.
18:49 – 23:24
07 · BCG's 10-20-70 rule
10% algorithm, 20% tech, 70% people and process. Most teams optimize the 10%.
23:24 – 24:04
08 · The correct sequence
Understand then clean then choose tools then build. Most people start at step 3.
24:04 – 26:28
09 · 48-Hour Shadow Audit
Every person tracks tasks live every 30 minutes for 2 business days. 3 columns: what, how long, done before this week?
26:28 – 31:47
10 · Time-Value Matrix
2x2 grid. High time / low value = automate. High time / high value = augment. Low time / low value = batch or stop. Low time / high value = protect.
31:47 – 34:43
11 · Scoring automation candidates
Multiply frequency x time cost x simplicity. Lead follow-up scores 504; rewriting SOPs scores 12. Build the 504 first.
34:43 – 38:02
12 · Clean data: the foundation
60-second client status test. Experian: 94% of businesses have inaccurate customer data. Gartner: poor data quality costs $13M/year average.
38:02 – 43:52
13 · Source of truth by revenue tier
Google Sheets under $5M, Airtable/Notion relational for $5-20M, Supabase/Postgres for $20M+. One place where the truth lives.
43:52 – 44:53
14 · Three rules for data hygiene
Standardize everything, mandatory fields, automated monthly audit. Without the audit, entropy wins in 60 days.
44:53 – 49:45
15 · The four-layer AI stack
Layer 1: Memory (data). Layer 2: Brain (LLMs swappable via API). Layer 3: Builder (Claude Code). Layer 4: Hands (specialized tools added only when backlog demands).
49:45 – 54:55
16 · MCP: the USB-C for your AI stack
Before MCP: 10 tools x 10 data sources = 100 custom integrations. With MCP: one connection per tool. Open-sourced Nov 2024, adopted by OpenAI and Google by March 2025. 97M monthly SDK downloads.
54:55 – 66:15
17 · Tool selection framework
Zapier/Make for simple one-off automations. Claude Code + MCP for real business logic. Always-on agents for proactive operations. Claude Code for custom anything that does not exist off shelf.
66:15 – 72:55
18 · Build 1: Lead response system
Claude Code in Cursor. Node.js backend. ICP scoring via Claude API. Gmail draft + Slack notification via MCP. Form demo with two test leads — one scored 2, one scored 8. Deployed to Railway.
72:55 – 77:13
19 · Deploying to Railway
Railway MCP server lets Claude Code push to cloud in one prompt. Public URL, 24/7 uptime, form submissions hit it directly.
77:13 – 82:55
20 · Build 2: Inbox triage agent
Classifies emails into urgent / client-lead / vendor-partner / newsletter. Drafts vendor replies. Sends one Slack digest. Scheduled daily. Human reviews all drafts.
82:55 – 97:11
21 · Managed Agents
Anthropic platform for production-grade agent hosting: credential vault, automatic error recovery, session transcripts, scheduled runs. Scheduler = phone reminder; managed agents = operations manager.
97:11 – 99:35
22 · Reactive vs. proactive systems
Reactive systems wait for triggers — deploy to Railway. Proactive systems initiate work on a schedule — deploy to Managed Agents.
99:35 – 102:58
23 · What else is buildable
Marketing content pipeline, sales enrichment with hiring signals, voice AI for after-hours calls, customer service handling 70% of tickets.
102:58 – 104:22
24 · Command center wrap-up
Green/yellow/red dashboard. 30-minute morning routine. Operator-to-owner shift made concrete.
Visual structure at a glance.
Named ideas worth stealing.
05:19
model
The 3 Eras of Business Operations
- Era 1: Manual (humans do everything, headcount scales revenue)
- Era 2: Tool-Augmented (siloed SaaS, team bridges the gaps)
- Era 3: Agentic Infrastructure (systems run the business, humans govern strategy)
A diagnostic model for identifying where a business currently sits and what the upgrade path looks like.
Steal for Discovery calls, positioning slides, or any content about AI readiness
07:59
list
The 5 Pillars of an AI-First Business
- Clean centralized data
- Intelligent workflows
- Connected systems
- Agentic operations
- Real-time visibility
The five components that must all be in place; missing any one means the business is AI-adjacent, not AI-first.
Steal for Audit frameworks, consulting deliverables, content about AI infrastructure
18:49
model
BCG 10-20-70 Rule
- 10% = the algorithm / AI model
- 20% = technology and infrastructure
- 70% = people and process
Most AI implementations fail because teams obsess over the 10% and ignore the 70%.
Steal for Reframing AI strategy conversations away from tool selection
24:04
concept
48-Hour Shadow Audit
Every team member logs tasks live every 30 minutes for two business days using three columns: what, how long, done before this week. Makes invisible repetitive work visible so automation targets become obvious.
Steal for Onboarding new clients, team productivity audits
26:28
model
Time-Value Matrix
- High time / Low value = Automate first
- High time / High value = Augment (AI handles prep, humans handle judgment)
- Low time / Low value = Batch or eliminate
- Low time / High value = Protect, do not touch
A 2x2 prioritization grid for deciding which processes to automate, augment, eliminate, or protect.
Steal for Any prioritization session, automation roadmap, team workshop
31:47
model
Automation Scoring Formula
Score each candidate on frequency (1-10), time cost (1-10), and simplicity (1-10). Multiply. Build the highest-scoring process first. Lead follow-up typically scores 500+; quarterly SOP rewrites score about 12.
Steal for Prioritization frameworks, deciding what to build first
44:53
model
Four-Layer AI Stack
- Layer 1: Memory — single source of truth
- Layer 2: Brain — LLMs via API, swappable
- Layer 3: Builder — Claude Code
- Layer 4: Hands — specialized tools added only when backlog demands
Architectural framework for building a composable AI stack. Each layer has a distinct role and strict sequence.
Steal for Stack design, client architecture proposals, content about AI infrastructure
97:11
model
Reactive vs. Proactive Deployment
- Reactive: sits and waits for a trigger, deploy to a hosting platform like Railway
- Proactive: wakes up on a schedule, initiates work, deploy to Managed Agents
Determines the right deployment architecture for any automation based on whether it responds to external events or initiates its own work.
Steal for System design decisions, client proposals
Lines you could clip.
03:17
"The gap is not adoption. Everyone has adopted. The gap is architecture."
03:20
"It is not about using AI tools. It is about redesigning how your business actually runs."
09:56
"You cannot fix plumbing by just adding more faucets."
19:59
"If your current process is messy, automating it just makes it messy at 10x the speed."
39:46
"Google Sheets — I am dead serious. You do not need Salesforce. You do not need a database."
47:46
"The point is not which tool you use. The point is that there is one place where the truth lives and everything else refers back to it."
How they asked for the click.
03:20
product
"If you are interested in getting a more hands-on approach, we have over 18,000 people building this stuff together in our free School community."
Multiple soft CTAs throughout. Webinar at 3:33 is the hardest pitch — unrecorded, gated scorecard bonus. Community links, newsletter, and agency booking link at end.
Word for word.
HOOK opening / re-engagementCTA the pitch
00:00HOOKI'm going to walk you through the entire process of making your business AI first from the ground up. So how to figure out what to automate, how to fix your data so things actually start working, and how to pick your tools. And then I'm going to sit down and build real systems on screen, so a lead response system that runs in under sixty seconds, even an AI agent that manages your inbox, so you can see exactly how this works in practice.
00:24HOOKNow, just recently, we helped a client completely rewire their business all with AI. So they were a service company, about $3,000,000 in revenue, 14 people on the team, and the lead response time, it went from over a day to under sixty seconds.
00:36Their team, they stopped spending twelve hours a week on manual data entry, and that all went to zero. Now their close rate, it jumped 35%. It's the same team, same service, completely different result.
00:47And here's what's also interesting about the AI tools that they were using, is that you have access to every single one of them right now. The difference, it was not the technology. It was just about how they actually put it together.
00:57So in this video, we're going to be covering marketing, we'll cover sales, operations, delivery, even finance, the whole business. Everything is going to be time stamped down below. So if you do want to jump around and if you're new to the channel, my name is Nick.
01:09I've spent the last two years helping companies implementing AI and helping hundreds of entrepreneurs start their own successful AI business. Now, I say all of that to show you that this is literally what we do every single day. In this video, it is going to be you if you've got a real business, you've probably messed around with ChatGPT or maybe even tried a chatbot, and it felt kind of
01:32okay. And that's not because AI doesn't work. It's because using AI tools and becoming AI first, they're two completely different things.
01:40And that distinction, it is what this whole video is going to be about, how you can start automating and scale the business. So let's get into it. So most businesses that we've worked with and that we have just seen, they're using AI completely backwards.
01:52I see all the time. You see a cool demo, something flashy, you sign up for that tool, you play around with it for a couple weeks, and then it just kind of dies.
02:01It doesn't help, it doesn't plug into anything, it doesn't connect to your CRM, it doesn't talk to your project management tool. It's just another tab on your browser, and then even more months go by, and you've got this Frankenstein kind of situation where you've got your CRM over here, you've got your project management tool over there, QuickBooks,
02:19it's doing its own thing. Google Drive with files everywhere. None of it talks to each other.
02:24You have data siloing. And your team, they're the glue. They're the one having to copy data from one place to another place,
02:32manually updating things and chasing down information across four different applications. And the numbers on this, they're actually kinda wild. So the average small business, they're running about 42 different software tools, and only about 28% of those are actually connected to each other.
02:47So your team, they're spending something like twelve hours a week just being the human bridge between your own systems. And on top of that, most companies, they're wasting around a $135,000 a year on software that nobody's really using.
03:01And that's not an AI problem. That's just a plumbing problem, and you cannot fix plumbing by just adding more faucets. Here's what's actually going on.
03:08McKinsey, they looked at this. They surveyed about 2,000 companies, and 88% of them, they're using AI somewhere in the business. That's almost everybody, but only about 1% have actually figured out how to make it work across their whole operation.
03:22So the gap is not adoption. Everyone's adopted. The gap is architecture.
03:27HOOKCTAIt's how you put the pieces together. And that's what nobody talks about because it's less sexy than just showing off a new chatbot. And becoming AI first, it's not about using AI tools.
03:36HOOKCTAAnd I need you to hear that. It's about redesigning how your business actually runs. So the systems, they handle the systematic stuff and you can only deal with the things that actually need you.
03:47HOOKCTAAnd that's the whole thesis of this video. And we're going to be making that real. Hey, really quick.
03:51HOOKCTAI just wanna mention on June 3 at 7PM Eastern, I'm doing a free live webinar. And basically, what I'm gonna do is I'm just going to walk you through the four AI agency offers that are actually working right now that either my agency or my students are actively selling. Now the whole point of this session is you watch it, you figure out which one of these four is going to be fitting you, and then you just go land your first client.
04:15HOOKCTASo if you don't show up, you completely miss it. Oh, and if you do show up live, I'm going to be giving you this thing that I made called the AI offer selection scorecard. It's basically how I would pick the right offer if I was starting from scratch today.
04:26HOOKCTASo look at me down below in the description. Just click it, grab your spot, and I'll see you on June 3, but let's get back into this. Now with that being said, I do wanna mention that if you are interested in getting a more hands on approach to all this and being a little bit more active, We've got over 18,000 people building this stuff together in our free school community and a free AI newsletter I recommend to sign up for.
04:46HOOKCTASo feel free to join. Link will be down below in the description, but let's start building out this framework. Now before we actually build anything, before we talk about the tool, before we talk about the AI models, before any of that, you need to first understand where you are right now.
04:59HOOKCTABecause if you don't know where you are starting from, you can't build a plan to get where you're trying to go. And what I found working with businesses over the last two years is that every single company, they fall into one of three different eras. And understanding which era that you are in right now, it is the single most important thing before you make any AI investment or any moves whatsoever.
05:19So let me just walk you through them. Now, error one, these are simply manual operations. Now
05:25error one, that's effectively where every business starts. So humans, they're doing everything. Your CRM, it's a spreadsheet.
05:33Your project management, it is email chains and sticky notes, and you track leads in your head, or maybe on a whiteboard somewhere,
05:42wherever. And in error one, the only way to actually start scaling is through increasing your head count. So if you want to do more work, you have to hire more people.
05:49If you want to respond to more leads, in that case, you need to hire more salespeople. So there's a direct and linear relationship between revenue and the number of humans who are inside of the building. Now very few people watching this, they are still in era one.
06:02But some of you, you might have departments that are still running like this, where, for example, like your sales team, it might be tracking deals inside of your spreadsheet, or your operations, maybe it is running off of any email threads.
06:15So even if the rest of the team has moved on, sometimes pockets of era one, they survive. And why that matters is because those pockets, they become bottlenecks. Now moving on to era two, this is where most of you are actually right now.
06:27I know this because this is where every business that we work with is starting from. So this is where you have software, you've invested in some tools, you've got your CRM in place, your project management, maybe it's like HubSpot, maybe GoHighLevel, maybe Salesforce, or maybe it's even ClickUp, Asana, Monday, whatever it is.
06:46You've got accounting in QuickBooks, you got Xero setup, you got Files in Google Drive, you got Dropbox, and it feels like progress. It feels like you've modernized. You're not going back to bare bones, so you're not going back to square one using sticky notes anymore.
06:59You've got actual real software. Now the trap of Era two is that every one of those tools, it is siloed.
07:05It's on an island. So they don't talk to each other and your team. Like, your humans, they are still the one moving data between them, all manually.
07:13If you think about it, like a lead comes in through your website form, somebody has to manually enter that and enter that into the CRM. And then when they actually become a client, someone has to create a project inside of the PM tool. And then when an invoice is due, they have to go create it inside of maybe QuickBooks, and the whole time like information is getting copied, it's getting pasted,
07:34and just reentered across three different or four different types of systems. And remember that stat that I mentioned, where 42 SaaS tools for the average small business?
07:44That is two in a nutshell. So you have the tools, but tools without any sort of integration, they're just expensive switching costs. So you're paying for software,
07:54but your team, they're still doing all the connective work by hand. And what actually makes Era two incredibly dangerous is that it feels like you're doing well. You've got the dashboards, you've got the notifications,
08:06you've got the data, but if I asked anyone on your team, hey, like what's the current status of client x, and their last interaction, and their open deliverables,
08:16and the next invoice date, and if it takes them more than sixty seconds and involves opening more than one system, that is how you know that you are in era two. Very simple way to deduce that. Now, era three, moving on to the agentic infrastructure.
08:30This is where we are actually going. This is where things are headed. This is the destination,
08:35and this is what it truly means to be AI first. So in era three, systems, these effectively run the business.
08:42So you have data flowing automatically between every tool in your stack, AI's handling any routine decisions, and the staff that's just role based, the stuff that doesn't require judgment, and humans.
08:55Like humans are governing the strategy, they handle exceptions, and they're managing the relationships that are truly mattering inside of the business. So let me just give you a very concrete example. So an era two, when a new lead is coming in, for example.
09:09So here, we have a lead. This person, let's say within 60 automatically, the system is going to read all of their information.
09:20So we have their info. We have our system above over here.
09:28Right? Okay. So the system, it reads all their information.
09:34It checks if they are matched to your ideal client profile. So this is kind of the golden thing that you're trying to get to. It makes sure it puts them through a filter and
09:46sees it through the ICP. It then can write a personalized response. It can send off I mean, it can do a range of different things.
09:53Like, there's all these different things that it is leading to, all these conditional
09:59marks. So it can send off that message. It can create the
10:04contact inside of CRM with a lead score, and it can alert your sales rep on Slack with just like a one line summary. Right? So no human actually has to touch that.
10:15And, I mean, it can run twenty four seven. It can run at 2AM on a Sunday, just as well as it runs at 10AM on a Sunday. But the data that actually backs this up, research, it's showing that responding to a lead within five minutes makes you 21 times more likely to qualify a lead compared to waiting thirty minutes.
10:31And responding within one minute, like, that's a 391% increase in conversion rates. But the average B2B lead response time right now, it's forty seven hours.
10:41It's not forty seven minutes. It's literally forty seven hours. So you guys can check out the source, I'll have that link down below in the description.
10:47But Era three, this isn't some theoretical future. Like, it's a competitive advantage that is literally available right now. There's no reason that you should not be trying to get to this point inside of your business right now and getting ahead of your competition.
11:00So the businesses that actually get there first, they're going to be eating the lunch of everyone who's still stuck in era two, stuck in era one, and this video, it is going to be getting you to era three. So that's the whole point. Make sure you stick around.
11:13And with that being said, there is a deeper shift that actually happens between all these eras, and I wanna call it out explicitly because this is the part that changes everything for business owners specifically. From era one, era two, you're effectively an operator. You're working inside of the business.
11:28You're touching the tasks. You're managing the fires. You're the person who actually knows where everything is and how everything works.
11:35And if you go on vacation for a week, like, things start to naturally break. And era three, you're the owner.
11:41You are working on the business. You review dashboards. You make strategic decisions.
11:45You handle the things that genuinely need your judgment and your relationships. And everything else, it runs without you. And I'm not saying this to be aspirational.
11:53I'm literally just saying this because I've seen it happen. I've seen business owners go from spending their entire Monday doing admin work, and catching up on emails, and updating CRM, and just reconciling invoices,
12:05to opening a single command center that shows them exactly what happened over the weekend, what needs their attention, and what's already been handled. And then Monday morning, it becomes a thirty minute strategic review instead of an eight hour catch up session. And that's the operator to owner shift.
12:20And everything in this video, it is designed to get you there. Alright. So enough talking.
12:24You know where you're going, era three. But what does an AI first business actually look like structurally? So what are the pieces that you need to have in place?
12:33So I've actually broken it down into five different pillars as you guys see on screen here. And every AI first business that I have seen, whether we have built it or someone else did or if they've done it internally and figured it out, it has all of these in place. And the whole thing, if you miss one, it will wobble.
12:50You'll run into issues. So let's walk through one by one. So pillar one, this is clean centralized data.
12:55So this is just having one single source of truth. There's no silos, and every system inside your business, it reads from, and it writes to the same core data.
13:05So if your CRM says that a client's name is Acme Corp, and your invoicing system says Acme Corporation, your automation, it treats those as two completely different companies.
13:16So clean data, it isn't just something that's like nice to have. Practically, it is essential to have.
13:22It's the foundation that everything else is going to be built on. So we're going to go deep on this in part four, but let's move on to pillar two. These are the intelligent workflows.
13:31So this is your business logic that's encoded into systems so that AI can execute decisions twenty four seven. It's just the rules of how your business operates, and it's written in a way that a machine can actually follow them. So hypothetically, if you're talking to your system, you can say like, if a lead matches these three criterias,
13:48send this email off. So if this, then that. Right?
13:52If a project hits 80% budget, then flag the manager. That's an intelligent workflow. Now pillar three, these are the connected systems.
14:00So within the connected systems, every tool inside of your stack, it is going to be talking to every other tool. So you have zero copy and pasting, you don't have to manage any of that. And when something happens in one system, every other system should be or it really will needs to know about it, and it gets updated automatically.
14:16So this is where something called MCP, model context protocol, this is where this is coming in. And we're gonna cover that in detail in part five. For now, just know that MCP, it's basically a USB c or just a USB for your AI stack.
14:29It's one universal connector that lets everything else talk to everything else. Okay. Pillar four, this is going to be your agentic
14:37operations. Now these are the AI agents that do not just follow a simple script. Instead they can reason, they can adapt to anything, and they act across multiple different systems.
14:49So the difference between a workflow and an agent, it's like the difference between a checklist and a smart employee. So the checklist, it does exactly what is written,
14:59and the smart employee reads the situation, it figures out the best approach, and we will be building both types in part six. For now, pillar five,
15:09real time visibility. So this is just a command center that shows you exactly what's happening at every single moment. So green, for example, this just means that everything is running, no issues.
15:18Yellow, this means that AI flags something, it needs a decision within a couple of hours. And red, this effectively means immediate human action, maybe a Slack alert triggered, and that's your daily routine.
15:31So you open the command center, you handle red and yellow, you scan the metrics, close your laptop. That's it. That's your day in operations.
15:38So you have these five pillars. You have clean data, intelligent workflows, connected systems, agentic operations, real time visibility.
15:46So if you miss any one of these, you're really not AI first. You're not doing and being as efficient or capitalizing on everything that you should be. You're rather AI adjacent.
15:55And AI adjacent, it doesn't really compound too well. It just costs money. So that right there, that's the foundation.
16:00You now understand the three eras where you probably sit right now, where you're trying to go, and what the structure of an AI first business actually looks like. But understanding it and building it, they are just two very different things.
16:12And that's what the rest of this video is about. And then the next section, we're going to talk about why most AI implementations are actually failing, and it's not the reason that you think.
16:21So from there, we're gonna do a full audit of your business to find the real opportunities, and then we're going to fix your data foundation and build your AI stack, and actually build real working systems using Plug Code, all step by step. And by the way, if you're a business owner looking to have us come in and implement any of this into your operation, then you can book in a call with our team to see how we can help you and your team.
16:39But, alright, let's talk about why most businesses actually fail with AI. In this next part, it might be a little uncomfortable, but it's the most important section of the entire video, so stick along. Alrighty.
16:49So if I'm just going to be as upfront and direct as possible, asking ourselves, like, why do most businesses fail with AI? So I need you to understand, most of them do fail. And this isn't one of those things where you just buy the right tool and it works out.
17:04Like, the actual numbers on this, like, they're pretty sobering. About 80% of AI projects, they never make it to production, and that's from Rand Corporation. And they found that's roughly double the failure rate of regular IT projects, and it gets worse.
17:17MIT, they did a study where they interviewed about a 150 companies, and they surveyed roughly 350 employees, and they found that 95% of generative AI pilots don't even deliver any measurable impact on the bottom line.
17:29That's 95%. So this isn't a some companies struggle situation, some others do well. This is the default outcome.
17:36So if you just do what most people do, you will most likely end up in that 80%, and I don't want that for you. So let's talk about what's actually going wrong and how you can account for that.
17:45So the number one thing that I see, and I've even seen this with our own clients before we got smarter about it, is businesses trying to do too much at once. So they get excited about AI, they see the possibilities, and they go, okay. Well, let's automate our customer service, and let's automate lead follow-up, and our reporting, and content creation, and onboarding, invoicing.
18:03And they try to tackle all of it at the same time. And what typically happens is none of it gets done properly. So you've got
18:0815 half built automations. None of them are reliable. Your team doesn't trust any of them.
18:14And three months later, everyone's just back to doing things manually, because the AI stuff, you kept breaking, and the businesses that actually win with AI, they do the complete opposite. They pick two, maybe three high impact areas. They go deep.
18:27They build it right. They prove that it actually works, and they like it, and prove it with real numbers, of course, and real time saved, and real money made, and then they expand from there. Now BCG,
18:37they have this framework that they call the ten twenty seventy rule, and I think it's the most important thing that anyone said about AI implementation. And that's only 10% of getting AI right is the algorithm. So the actual AI model.
18:49Now another 20%, it is the technology. It's the infrastructure.
18:53And 70, it's the people in the process. 70%.
18:57So that means if you spend all your time picking the perfect AI tool, and none of your time actually fixing your processes and getting your team on board, you're just optimizing for 10% of that problem. And that's why most implementations fail. It's because they start with the tool, and they should instead be starting with the business.
19:13Alright. So this next one, I need you to really sit with this for just a second. So if your current process, if it's messy or automating it just isn't fixing it, it makes it messy at 10 x the speed.
19:24So you're just creating faster chaos. And I'll actually give you an example. Let's say your lead follow-up process right now is kind of all over the place.
19:32So let's say sometimes someone responds in an hour to somebody, and sometimes it even takes two days to get back to a customer, whatever it is. There's no consistent template is what I'm getting at. And nobody really knows who's supposed to follow-up with which leads.
19:45It's just whoever gets to it. And now you come in and you automate that process with AI. So what happens is the AI, well, it naturally fires off inconsistent responses at lightning speed.
19:55It follows up with leads that should have gone to a different representative. It sends messages that don't really match what the salesperson already said, and now instead of your process being slow and messy, it's fast and messy, which is actually worse because now you're actively knowing potential customers at scale. So before you automate anything, like, the underlying process, it has to be clear, and here's my test for that.
20:15So I call this the new hire test. So could a brand new employee follow your current process from a written document with no help from anyone? And get it right.
20:24So if the answer is no, that process is not ready for AI, Because AI, it's basically just a new hire that follows instructions literally. And if your instructions aren't clear, then the AI will do exactly what a confused new hire would
20:37really be doing. They guess. They get it wrong.
20:40They create problems. So before you build anything, you document the process. You have to make it very clear.
20:45You have to make it step by step, and then you could actually automate it. And that order is one of the most important things. Now moving on to the tools versus the infrastructure.
20:54So we actually talked about this in part one, but I want to just drive it home here because it's the mindset piece that actually trips people up the most. So buying an AI tool, it is not becoming AI first. That's really just like buying a treadmill and calling yourself an athlete.
21:10And having a gym membership, it doesn't make you fit. But instead, using the gym makes you fit, and using it with a plan makes you fit efficiently. It's literally the same thing with AI.
21:20Having JatGPT on your browser does not make you AI first, or having a chatbot on your website does not make you AI first. Like, those are tools. They help one person do one task a little bit faster, a little bit better.
21:32Infrastructure, it's inherently different. So infrastructure means the system handles it end to end without a person involved at all.
21:39So again, like a tool is having like your sales rep use AI to write follow-up emails faster. Infrastructure is looking at, like, every new lead, they get a personalized response in under sixty seconds.
21:51The CRM, it gets updated. The rep, they get a Slack notification with a summary, and a follow-up sequence starts all automatically twenty four seven without anybody having to touch it. Now one, it saves your rep ten minutes.
22:03The other, it removes the task from their plate entirely, and it does it better than they could and at any hour of the day. And that is the difference right there. And once you actually start thinking in terms of infrastructure instead of tools, everything, and I do mean everything about your
22:18approach with AI, it completely changes. So here's the last piece of the mindset shift, and this one is at least what I tried to make as practical
22:27as it can be. So there is a correct order to actually doing this, and most people they get it completely backwards. So the right sequence, it's understanding your business first.
22:36So you have to figure out where the real opportunities are, and it's not based on where you think they are. You have to go where they actually are. That is step number one.
22:44Step two, cleaning your data, getting your information organized, because nothing that you build on, it's not going to work if the data feeding it is just a complete mess. Now step three, that's choosing your tools, and not doing so before you know what you're building, it has to be after. And step four, that's building.
23:00So most people, they start at step three, they see a cool tool, they build it, then they try to figure out what to do with it, and that's just backwards. That's how you end up with 42 different SaaS applications, and none of them are connected. Now the rest of this video, it follows that exact sequence.
23:15And then next up, we're gonna audit your business and find the real opportunities, and then data, and then tools, and then you're doing the building. You're doing that in that Now just a quick note, if you're watching this and you're thinking like, okay.
23:26I get the logic, but I'd rather just have someone do this for me. That's exactly what we do at Reprise dot ai. We run an exploration phase where we come in, audit your business, we will map opportunities, build infrastructure with you.
23:37So if you are interested in that, then the link will be down below in the description if that's more your speed. But everything I'm teaching in this video is the exact same process we use with our clients. So either way, you're getting a real playbook.
23:48And if you wanna stay plugged in beyond this video, I do break down AI implementation strategies and new tools and real builds inside of our weekly newsletter, AI Core. So link will be down below in the description. Alright.
23:59Now let's actually look under the hood of your business, because this next section is where we audit everything and find the real money. Alright. So we just spent a good chunk of time on all the mindset
24:08stuff, the eras, the pillars, you know, why most AI implementations actually fail. All of that matters, you need it before anything else. Alright.
24:15So this is what I call a forty eight hour shadow. So within this, your first step is for about two full business days, maybe it's Monday and Tuesday, whatever your next normal workdays are, every person on your team, including you, they should be keeping a simple running log. So you're just tracking three things.
24:29So what did I just do? How long did it take? And have I done this exact same thing in the past week?
24:34Like, that is it. You have it be about three columns, two days, and the keyword here is, like, running.
24:40So you do this live as the work actually happens, not at the end of the day when you try to reconstruct your morning from memory. Because you probably won't remember. Nobody remembers.
24:49You think you spent twenty minutes on email, but it was actually an hour and a half. You think that report took thirty minutes, but you got interrupted twice, and it ate your whole afternoon. So set a time on your phone, have your team do this every thirty minutes.
25:01It'll go off. You write down what you just did. It literally takes fifteen seconds each time.
25:06So there's this massive gap between what people think they spend time on and what they actually spend time on. In Asana, it actually found that employees spend about 62% of their workday on repetitive coordination tasks. It's not the work that they were even hired to do, but the work around the work.
25:22It's beating around the bush to get to the work. So scheduling, searching for files, updating statuses, moving data between tools.
25:29Now Process Maker, they track this even more specifically. So the average enterprise employee, they do over a thousand copy paste actions per week.
25:36A thousand. Like, across a 20 person operation team, that's over a million copy paste a year, and nobody sees this. It doesn't show up on a report.
25:44Nobody puts, I spent forty minutes finding a document to go drive on their to do list. It's invisible. It just happens.
25:50And invisible work is where the biggest automation opportunities actually hide. And the forty eight hour log, that is going to make this visible.
25:57Like, that's all this exercise is doing. It takes the invisible stuff, and it puts it on paper so you can actually see what and where your business is running on. Now, I want you to actually do this.
26:06So if you need to pause the video, open your calendar, set a reminder for your next day to start the log. And I'm not being dramatic.
26:14Like, this one exercise done properly is worth more than any AI tool that you can buy because it tells you what to actually start building. Now moving on to the time value matrix. So you've done the forty eight hours, you've got your list, it's probably a really messy one, that's fine,
26:29but now we need to sort it. And the way that you need to sort it, it is dead simple. You just have to take every task on that list and ask two questions.
26:36So one, how much time does it seat? Two, how much does it actually matter?
26:42So from here, you can plot everything on a two by two grid just like this. So you can see, you have your time spent on one axis. You have your strategic value on this other one.
26:53So the top left is, of course, it takes a lot of time, creates very little value. You have your data entry, reformatting reports,
27:01copying information between systems, sending the same follow-up email for the hundredth time, updating spreadsheets that three people maintain differently. In this quadrant, it is your gold mine.
27:11Like, these tests are just begging to be automated. So they're high volume, they're low judgment, and your team hates doing them. So that's where you start.
27:19On the top right, this takes a lot of time, but it's genuinely valuable work. You have your sales conversations, your client strategy, any complex proposals,
27:28creative problem solving. You don't automate these, you augment them. So what I mean by that is you let AI actually handle the research, all the preparation work, the first draft, and you just let your people focus on the parts that actually need their brain.
27:41Now the bottom left, this doesn't take much time. It doesn't also matter much, but you batch these or you just stop doing them. So you don't want to waste an automation build or spending your valuable resources
27:52and effort on something that takes ten minutes a week. And then lastly, the bottom right, these are quick but extremely valuable. So maybe it's a five minute client check-in that just builds loyalty, or maybe a fast call that saves a deal.
28:04You leave this completely alone. They work right now. And when most business owners actually do this for the first time, they're very often surprised by how much of their team's week is just sitting in that top left quadrant.
28:16And the research, it backs this up. Like the average worker, they spend 51% of their workday on tasks of little to no value.
28:22In Asana's data, it shows about 60% of employees' time goes to just work about work. So the coordination, the status updates, and searching for information. So it's not even the skilled work that they were hired to do.
28:33McKinsey, they even found a similar breakdown, where 28% of the average work week, it just goes to email, and another 19% to searching and gathering information. And that's where the hours are actually hiding, and now you can see them.
28:46So now you guys know which processes to actually target, but here's where most people make a mistake that ultimately kills the whole project before it even starts. And that's where they look at something like, maybe it's a lead follow-up, and they try to automate lead follow-up, of course.
28:59So they go looking for an AI tool that does the lead follow-up, and of course, nothing does exactly what they need because that system, that process, it isn't a task. It's just a label. You can't automate a label.
29:12What you can automate is the steps. So let's take lead follow-up and actually just break that down. So what does a person on your team do?
29:21Physically, step by step, when they follow-up with a new lead. Okay.
29:25So they open the CRM, that's gonna be step number one. They filter for leads from the last twenty four hours, that's gonna be step two. They click into each lead, and they check.
29:34Like, okay, does this person match our ICP? And then step three, they open the intake form, they read what the person submitted. That's gonna be step four.
29:43They draft a personalized email based on what they read. And then step five, they send it. Step six, they go back to the CRM.
29:50They log that they sent it. And then step seven, they now set a follow-up reminder for about three days out.
29:56That's step eight. So you have eight steps, and now if you look at each one on its own, opening the CRM.
30:02Okay. So does that require any human judgment? Let's say no.
30:07What about filtering leads? Does that require human judgment? K.
30:10Let's say no for that. Checking ICP match. So if you have clear criteria written down, that's gonna be a rule, not necessarily a judgment call.
30:18And then reading the intake form. So an AI, it can read a form, drafting a personalized email based on what's in the form, AI is actually quite good at this, and then sending it off. That's automated.
30:28It can be done. And logging it. That can also be automated.
30:32Setting a reminder. Also automated. So out of eight steps, maybe one needs a human to glance at it in the beginning.
30:38Everything else, that's gonna be role following, and that's the pattern pretty much everywhere. When you do this exercise across your whole business, and you break every process into the smallest steps, and mark which ones require judgment versus which ones are just following rules, you'll find that somewhere between 60 to 80% of most business processes, they're just pure role following.
30:57There's not much creativity needed, no judgment needed, and just somebody doing the same thing the same way every single time. And that's your actual automation opportunity. So it's not just like automate marketing or something as bare bones and generic as that.
31:10It's not automating sales, but instead automating the 47 specific steps inside those categories that do not need a human. And the other 20 to 40%, like the judgment calls, the relationship moments, the creative decisions, like that's where you want your people spending all their time, and that's where they're good at.
31:28That's what moves the needle for you. So now with this, you've got a list of all your automatable processes. They're broken into steps, but the question is, like, where to actually start?
31:37And this matters than more people think, because your first automation needs to be visible. It needs to be a very undeniable win because that first win is what gets your team bought in. It gets them on board.
31:48It's what gets your business partner to actually approve the next project. So if your first build saves twelve minutes a week and nobody notices, that's a terrible start. You just wasted so much time doing that.
31:58You need something where people are actually going like, okay, this is amazing. It's helping my life. It's helping me be more efficient.
32:03We are seemingly getting up increasing our bottom line, or whatever it may be. So here's how I actually pick. We score every candidate on three different things, and we multiply the scores.
32:14So first one is the frequency. So how often does this happen? Is it a daily process that gets an eight, or gets a nine, or gets a 10?
32:22And then monthly, that is getting a two or a three. Now the time cost.
32:26So how many hours per week does it actually eat across your team? So you need to be very honest with this and include the invisible time that you found in the forty eight hour log. Next up is the simplicity.
32:37So how clear and rule based is the logic actually? So could a new hire follow a written checklist and get it right?
32:45If yes, then that's an eight, it's a nine, it's a 10. And if it involves a lot of exceptions and judgment calls, then it's probably a three or four. So you multiply the three numbers, and then from there, the highest score, that's what's going to be winning.
32:58So for example, we have lead follow-up. Like, this happens every single day. Frequency's about a nine.
33:04It takes about five or six hours a week across the sales team, so the time cost is a seven. Now the steps, they're clear, they're rule based, and the simplicity, it's about an eight. So nine times seven times eight,
33:17that's 504. Now if you compare that to something like rewriting your SOPs, that happens what? Maybe once a quarter?
33:22So the frequency, that's about a two. And this takes ten hours when it happens, but averaged out, it's maybe one hour a week. So the time cost, this is a two, and it's highly judgment based.
33:34So the simplicity, it's a three. So two times two times three.
33:39Of course, if you do some simple math, that's 12. So 504 versus 12.
33:44So the lead follow-up, it wins by a mile, and it'll be live in one to two weeks. Like, that's gonna be your first build.
33:50So don't overthink this. It may not always be as simple as that, but the whole point of the scoring is to just stop you from going after the sexy project and get you to build the one that actually delivers fast, obvious results.
34:02So the momentum, is going to matter much more than your ambition or what you may think at this stage. Alright.
34:08So at this point, you know exactly what to automate and in what order. You've got your list. You've got your scores.
34:14You're ready to go. And I know, like, right now, you probably wanna jump in the tools, you wanna start building. I get that.
34:20But remember what we said in part two, there is a correct sequence to do things, and if you understand your business first, like, okay, well, obviously, you just did that. You clean your data second, and then you go to the tools third, and then you build fourth. So we're at step two, your data.
34:35And this one, it's kinda boring, but it's very critical because if your data is a mess, and I promise you some of it probably is, then nothing that you build on top of it's going to work right. So garbage in, you get garbage out.
34:47And that's not a cliche. Truly, it's not a cliche. It's literally the number one reason that automations break, so let's make sure we fix that first.
34:54But I wanna try something. So before I even explain anything about data, I want you to do a test. So
35:00right now, or as soon as you finish this section, go ask anyone on your team this question. Pick a client,
35:07literally any client, and ask your team member. Like, what's the current status of that client? What is their last interaction with us?
35:14Whether that's their open deliverables or the next invoice date. And the time how long it actually takes them to answer. And if they can pull that answer up in under sixty seconds from a single place, you're in great shape.
35:24But if it takes them two minutes, three minutes, five minutes, ten minutes, whatever it may be. If they have to open the CRM to check the less interaction and jump to the PM tool, find the deliverables, and go to QuickBooks,
35:34you know, you have a data foundation problem. And everything that you try to automate on top of that foundation, it is going to be unreliable, and it is going to be very shaky. So that's the sixty second test.
35:45And most businesses, they fail at it very poorly. So let me just really quick tell you what happens when you try to build AI systems on top of messy data. Because none of this is theoretical.
35:54I've watched this happen. This is very common. You build an automation that's supposed to send a follow-up email to every new client after their onboarding call.
36:02I mean, ideally, that's pretty smart. Right? So that should save your team maybe three hours a week, let's say.
36:08Except the automation, it pulls the client's name from your CRM, And your CRM, it has the client listed as Acme Corp. But your invoice system, it has them as Acme Corporation. And your project management tool, it has them as Acme Co.
36:21So now the automation, it either can't match the records and the email never goes out, or it creates a duplicate entry because it thinks that these are three different clients. So your team finds out two weeks later, you know, somebody says like, hey, never got that onboarding email. Now you've got a mess that just takes longer to clean up than it would've taken to just send the email manually in the first place.
36:39And that's just a mild example. Like, I've seen much worst instances where I've seen automation send proposals to the wrong contact because the CRM, it had two records for the same person, one created by marketing, one created by sales with different email addresses.
36:52So one was the person's old email for a job they left two years ago. Automation picked that one. Proposal goes into the void.
37:00Sales rep has no no idea about that. Deal goes cold. They lost money.
37:04So this stuff sounds small, but a name formatted differently here and then a duplicate record there, you know, you start laying on automation on top of that, and every small data issue becomes an underlying system level failure. And it happens very fast. Trust me.
37:18Like, the data on this Experian, they found that 94% of businesses suspect their customer data is inaccurate.
37:25So it's not a small problem. Gartner, they estimate that organizations with poor data quality, it loses about $13,000,000
37:32per year on average from the downstream effects, and that's not the cost of fixing it. That's just the cost of leaving it broken. And the thing that actually makes us really insidious is that it's invisible.
37:41Nobody wakes up and says, our data's a mess. Like, that's why our automation broke. Like, they just assume that AI doesn't work for them, or they tried it.
37:49They blame the tool. They blame the agency that implemented it. They blame the model.
37:53Nine times out of 10, like, the tool and the model, it's fine. It's just the data feeding them. It's just complete garbage.
37:58And, again, garbage in, garbage out. I used this at the end of part three, and it sounds like a cliche, but I want you to understand what that actually means in practice. So it means if you feed an AI system dirty data,
38:10it doesn't know the data is dirty, and it doesn't say, hey, this looks wrong. Let me double check that. It just uses it.
38:16So it's gonna use it confidently, and it produces confidently wrong results. Who would've guessed?
38:21So you can't automate what you cannot trust, and that's the rule. Make sure you follow that. So how do you actually fix this?
38:27Well, the concept's actually very simple even though the execution, it will take some work. So you need one place where your core business data is actually living. You need one source of truth.
38:36Now every other tool in your stack is going to be ready from it, and when something changes, it is going to be changing in that one place and ripple out everywhere else. So what that actually looks like, it depends on how big your business is and where you are actually at.
38:49And I'm gonna be really specific here because vague advice, it'll probably be very useless to any of you guys. So if you're under $5,000,000 in revenue, and this might surprise some of you, Google Sheets. I'm dead serious.
39:01You don't need Salesforce. You don't need a database. You need just one clean, well structured Google Sheet that serves as your master record.
39:08One sheet for clients, one for leads, one for projects. You could have one for financials. Like, every tool that you use, your CRM, your project management tool, your invoicing software, like, that either reads from these sheets or it writes back to them.
39:19Like, no tool gets to keep its own private copy of the data that nobody else can see. Like, is it Glamorous? It's not.
39:26But does it work? Absolutely. I've seen businesses doing $34,000,000
39:31a year running their entire data layer on Google Sheets with automation just connecting everything. And it's clean, it's reliable, and when they're ready to upgrade, like the data is already structured,
39:41it's ready to move. It's very simple, it's very easy to follow. So if you're anywhere from five to 20,000,000, this is where you can use something like Airtable.
39:49You can use Notion with a relational structure. So this means that your clients, they're linked to their projects, which are linked to their invoices, which are linked to their contacts. So it's all just connected in one place, and it's not sitting in separate tools.
40:01Just hoping somebody remembered to update both of them. So the key here is relational. So your data, it's not just being stored, but it's being linked.
40:10So when you look at a client, you can see every project, every invoice, every interaction. It's all from one record. And that's what makes AI actually work, because when the AI needs to answer a question about a client, it can pull from one connected dataset instead of trying to stitch together information from four different tools.
40:24So if you're at 20,000,000 plus, like, probably want something a little bit more complex, you can use Superbase, you can even use Postgres, HubSpot, all that stuff. So, like, you need real time API access.
40:35So every tool in your stack, it queries one database. And this is where you get into the real infrastructure. But honestly, most people watching this don't need to be here yet.
40:43Get the Google Sheets, your Airtable layer just working first, prove the systems, and and then you can migrate to a database when you actually outgrow it. But the point isn't which tool you use. The point is that there's one place where the truth lives, and everything else refers back to this.
40:58So with that being said, you now have your source of truth set up. You now need to keep it clean, because data hygiene, it's not a one time project. It's an ongoing system, actually.
41:09So you clean it once, and if you don't have rules, it's dirty again in sixty days. So it's three rules for nonnegotiable. Rule one, you just have to standardize everything.
41:19So every name, every label, every format, it needs to be consistent across every single tool. So if a client is Acme core in in the CRM, they're Acme core in invoicing and project management,
41:30everywhere. So it's not Acme Corporation, not Acme Corp, not Acme. It has to be the same exact string.
41:36Now I know this sounds pedantic, but it actually matters enormously. So your automations, they need to match records by name. And if the names don't match exactly, then the automations, they'll likely either break or create duplicates, unless you're using some sort of agentic system.
41:50So you have to pick one format, you have to document it, and you have to enforce it. And the same, it goes for dates, it goes for phone numbers, and even addresses, like all of it. You pick a format, stick to it.
42:01There's no exceptions with that. Rule two, these are mandatory fields. So every lead, every client, every project, they must have certain fields filled in before it can move forward.
42:11So you have to define what those fields are. Like the email, the phone number, the company name, the source, whatever matters for your business, and nothing with this can progress until they're actually populated.
42:23So this is one of those things that feels annoying for about two weeks, and then it just becomes invisible. So your team, they're just getting used to filling in the required fields because the system, they won't let them skip it. Six months later, you have a database where every record is complete, That's and incredibly powerful for when you start building automation because you never have to worry about anything missing data causing a failure.
42:43And most CRMs, they let you set these required fields natively. So if yours does not, you have to build a simple check at the workflow level. So if field x is empty, you should not be proceeding with that.
42:54Moving on to rule three. These are the automated monthly audits. So you could just set up a simple workflow, and this takes maybe about thirty minutes to build.
43:03And that runs just once a month, and it flags any records that are simply incomplete, duplicated, or even conflicting.
43:11And it sends you a report, so you or someone in your team can just spend an hour just cleaning it up. Like, that's it. One hour a month, your data stays clean, and without that monthly audit, like, Entropy, it'll win.
43:23And people get sloppy. Fields get skipped. Duplicates, they'll creep in, and then six months later, you're back to the mess that you just started with.
43:30So three rules, standardize them, require fields, audit monthly. If you do nothing else from this section, make sure you do those three things. Alright.
43:37So here's where we are. You've audited your business. You know what to automate.
43:41You've cleaned your data. You set up a source of truth, and you've got rules to keep it clean. Now, finally, we'll talk about the tools, and I know some of you have been waiting for this since minute one.
43:50Like, what tools I use? What's the stack? How do I fit it all together?
43:54That's gonna be next, and it's going to go faster than you think because when you've done the work that we just covered, when you actually know, like, what you need before you go shopping, picking tools becomes obvious instead of overwhelming, I will say. So let's start building your stack. So we've spent the first four sections on the purpose and understanding where you are, fixing your mindset,
44:13auditing your business, cleaning your data. Like all of that, it's completely necessary. And if you skipped any of it, seriously, you need to go back and watch that.
44:21Because what we're about to build only works if those foundations are solid. But now we have to talk about the tools, and now we build the stack. And here's an idea to understand before we actually dive in, is that there's one rule that governs everything in this section.
44:34So that's the architecture first rule. Now, every tool that you add to your stack, it should be serving as your automation backlog. Like, there's no other way around this.
44:44I see this mistake constantly. Somebody, they're watching a YouTube video about a new cool AI tool, a new demo. They buy it, and they try to figure out what to actually do with it, and that's inherently backwards.
44:56Like, shopping without having a list even. And it's how you actually end up with that Frankenstein, that siloed data,
45:03which is what we talked about in part one. So you did the audit in part three, you scored your processes, you know exactly what you need to build first in that backlog, that scored list, that's your shopping list.
45:14You don't pick a tool and then look for a problem. You have to look at the problem first, and then pick the tool that actually solves and accounts for it. And that sounds obvious, right, but in practice, nobody does that.
45:24So here are the four layers of an AI stack. So let me show you how this actually fits together because it's not just a pile of tools. It's actually an architecture here.
45:34So there's four layers. Each one, it serves a very specific purpose. And with this order, it matters.
45:41So layer one, this is the memory. This is your data layer. So this is what we just built in part four.
45:47So you're a single source of truth. Now Google Sheets, if you're under 5,000,000, add a table or Notion with relational structure if you're about five to 20,000,000, soup base or Postgres if you're bigger than that.
45:58That's the foundation. Everything else, it sits on top of this. So if a layer is messy, nothing above it is going to work right, and we already covered this in detail, I'm not gonna be repeating it again, but just know that on this diagram, everything starts here.
46:10Have data at the bottom. Now layer two, you have the brain. These are your LLMs.
46:15So you have Claw, GPT, Gemini. Like, these are the language models that actually do the thinking for you. So they read your data.
46:23They understand the context. They make decisions, and they generate content. Right?
46:27So here's my advice on this layer, and it might be different from what you've heard anywhere else, is you can't get locked to just one model. So the AI space, it moves incredibly fast.
46:37And Claude, is best at coding and long form reasoning right now. In GPT, it's strong at certain creative tasks.
46:44But Gemini, it has advantages with Google ecosystem integration. So six months from now, landscape might look completely different.
46:50So you build your system so the brain is actually gonna be swappable. And you can use APIs, not proprietary platforms that just lock you in.
46:57Because that way, when a better model does come out, and it 100% will, you switch the brain without rebuilding the body. Now layer three. This is what I call the builder, and these are just your agentic coding tools.
47:09So this is the layer that really changes everything for businesses like yours specifically, and I mean that. Like, this layer is why you're watching this video right now instead of hiring a development team.
47:19So Cloud Code and tools like Cursor, like, are not auto complete. They're not suggesting the next line of code. They're actually full on AI agents that can plan, that can build,
47:28test, and even ship real working software. So you describe what you need to play in English. So for example, you could say, I need a system that takes every new form submission, checks if that person matches our ICP, sends them a personalized email within sixty seconds, and creates a contact in our CRM, and alerts my sales rep on Slack, and then the tool bill does it.
47:46So to give you a sense of how real this is, so just to give you a sense of this, like, Cloud Code, they just launched publicly in May 25. And by November 25, six months later, it hit $1,000,000,000 in annual revenue. Now that's faster than ChatGPT, faster than Slack, faster than Zoom.
48:00And by early twenty six, it's estimated at over 2,500,000,000. And over 500 companies are now spending more than $1,000,000 a year on Anthropic products. Like, This isn't a toy.
48:09This is real infrastructure. And the reason it's growing that fast is because it solves a real problem. Businesses like yours,
48:14they need custom systems built, and they don't have engineers or the teams to do it. And Claude Code, it is the engineering team. This is monumental change in the space and
48:24really everything. Like, we're gonna use this to build real systems from scratch in part six, but just know that, and I'll walk you guys through it step by step.
48:32So layer four, this is the hands, and these are your specialized tools. So these are the things that actually do specific jobs. So things like GoHighLevel or HubSpot for CRM.
48:42You have instantly dot ai for cold outreach. You have Fireflies for meeting intelligence. You have Bland.
48:47You have Vappy for voice AI, and your accounting software, and your project management tool. So the critical thing about this layer, and this is where the architecture first rule is really kicking in, is that you only add a tool
49:00when your backlog is demanding it. So it's not because somebody recommended it or just said so. It's not because it was on sale.
49:06It's because you have a specific automation that actually requires this specific capability, and there's nothing else in your stack that can handle that. Now most businesses, they have about 20 different tools in this layer when they really need just a few. That's because they didn't have the first three layers in place, so they kept building specialized tools to compensate for missing infrastructure.
49:23But when your data layer, your AI layer, and your builder layer are actually very solid, you need far fewer specialized tools. Now, let's move on to the connective tissue,
49:33which is MCP. So you've got four layers, but the layers themselves, they just won't do anything if they can't talk to each other.
49:41And this is where something that I mentioned earlier, MCP comes in. And I need you to understand this because it is quietly the most important development in AI infrastructure in the last two years. MCP, model context protocol.
49:53Easiest way to understand this is if you remember early in the February, every phone had a different charger. Motorola had one shape, Nokia had another, Samsung had a third.
50:04And you'd go to your friend's house, nobody had the right cable for your phone. Right? So that was a mess.
50:08And then USB C came along. One cable, every device. Doesn't matter if it's a phone, doesn't matter if it's a laptop, a tablet, a camera.
50:15You have one connector, it all works. MCP, as I mentioned earlier, it's a USB c for your AI stack.
50:21So before MCP, if you wanted your AI to connect to your CRM, you needed a custom integration for that CRM. If you wanted it to connect to Google Drive, you needed another custom integration. Slack, another.
50:31Calendar, you need another one. So if you had 10 AI tools and 10 data sources, you potentially needed a 100 different integrations. That's the old way.
50:39That's the mess. But with MCP, you just build one connection per tool. So your CRM, it speaks MCP.
50:45Your Google Drive, it speaks MCP. Same as Slack. And any AI agent that speaks MCP can talk to all of them through one universal interface.
50:54One standard. There's no custom glue code. So Anthropic, they're the ones who created this protocol.
50:58They open sourced it in about November 2024. And by March 2025, OpenAI adopted it.
51:04Google DeepMind adopted it. Microsoft, and within a year, it went from one company's experiment to the industry standard. And now they have over 97,000,000
51:14monthly SDK downloads and thousands of MCP servers for different tools and services. And in Thropic, they donated it to the Linux Foundation under the Agencik AI Foundation to make it vendor neutral. So what that actually means for you practically is that when you build an automation using Cloud Code, let's say a system that reads your emails or checks your CRM,
51:31drafts a response, logs it, all of those connections, they happen through MCP. One standard protocol in the AI, it connects to Gmail through MCP. So it connects to the CRM through MCP, connects to Slack through MCP.
51:43You don't need a developer to build custom API integrations for each tool. You don't need the API documentation and all that stuff. You just plug it into the MCP server for that tool, and it will start working for you.
51:54So now the systems that actually can run your business and that can handle any complex logic that need proper error handling, like quad code plus MCP, it's the new standard. It's not any nine.
52:05It's not make.com. It's not Zapier. And that's because you own the code.
52:09You control the logic. It's not locked in inside someone else's platform. And when something breaks, like, you can see exactly why and fix it instead of staring at a visual workflow
52:18and trying to figure out, like, why the node is failing. Cloud code plus MCPA. That is the standard.
52:24You have to be using that if you wanna be as efficient as possible. So it's not to say you shouldn't use any of that anymore. I still use it internally.
52:31It's very practical for small things or any systems that you just want up and running. And I know that might be a lot to take in if this is the first time you're hearing about MCP or code, like, that's okay. You don't need to understand all the technical details.
52:45What you do need to understand is that MCP means your AI systems can talk to any tool in your business through one universal standard, and that changes everything about how fast and how reliably you can actually build. So let me give you a simple decision framework for where each tool is going to be fitting because I get asked this all the time, like, should I use Zapier?
53:06Should I use Cloud Code for this? So here's how you actually think about that. So for any quick one off automations, use any none or Zapier or Make.
53:13So if it's like a when x happens in tool a, do y in tool b. And the logic is very simple.
53:20You just use Zapier. You don't have to overthink it. Just set it up.
53:24Takes five minutes. That's it. Now for any business workflow automations,
53:28Claude code plus MCP. So if there's real logic involved, if there's a lot of conditions, if there's error handling needed, if the system needs to make decisions, if you want to own the code and not be logged into a platform, like, this is where the real infrastructure lives. Now for any always on AI agents,
53:43this is a newer category, but it's growing very fast. So tools like OpenClaw, it's open source.
53:49It gives you an AI agent that runs twenty four seven and proactively manages operations. And it doesn't just wait for a trigger. Watches your business and notices things.
53:58It takes action. So you think of it as like a virtual operations manager that never sleeps.
54:02And then we have custom AI infrastructure. So Claude code. Anything that doesn't exist off the shelf.
54:08So if you need something unique to your business, something that no SaaS product does, Claude code, it builds it for you. Like, this is the equivalent of having a full development team on call, except the build takes
54:19minutes instead of hours or even months. Like, the thing with that, so you probably need all four of these at different points. Some are gonna be better for other things.
54:28Zapier for the simple stuff, or NNA nine for the simple stuff, Cloud Code for anything, like, real, and an agent for the ongoing stuff, and custom builds for the unique stuff. It's not either or, it's a toolkit.
54:40And you pick the right tool for the right job. But overall, Claude Code's probably going to be the best bet for everything. That's what I use pretty much for everything.
54:47I have Claude Code build my any of that automations, and really anything else. I use it for probably 90%
54:53of everything that I do. Alright. So moving on, here's how to actually start building your AI infrastructure.
54:59Now everything up until this point, it was just prep for this piece. So the framework, the audit, the data cleanup, the stack, like, all of that was getting you ready for this moment. Now we actually build something.
55:08And I wanna be upfront about all this. Like, I've made a ton of videos about Cloud Code on this channel, like deep dives, tutorials. If you wanna go way deeper on Cloud Code specifically,
55:17go check those out after this. You can get them inside of my free scroll community for a consolidated playlist and all that stuff, and you can discuss with other people inside of that community. But what we're doing here is just showing you the method.
55:27So we're gonna be building two real systems from scratch so you can see how this works and understand that you can do this for your own business. And these are just gonna be simple builds, like foundational stuff, but the approach, it is the same inherently whether you build a lead response system or a full operations dashboard.
55:41So let's start with what Claude code actually is because there's a lot of confusion about it and whether you should be using it or not. So Claude code is not a chatbot. For example, when you use Chaijibouti or you use Claude in the browser, and you type a question, you get a response.
55:55Like, that's in chatbot. Claude code is not that. Claude code is also not auto complete.
56:00So, like, you might have seen tools like GitHub Copilot, where you're writing code and it suggests the next line. Right?
56:05Claude code's not that either. The best way that I can describe this is that Claude code is an AI employee that builds software for you. So you tell it what you need in plain English.
56:14Like, it can't be in code, can't be in any technical language. You just say something like, I'm the system that takes every new form submission from my website, checks if that person matches their ideal customer profile, sends them a personalized email within sixty seconds, then creates a contact in our CRM with a lead score, and then sends my sales rep a Slack notification with a summary of who this person is.
56:35Now, Cloud Code goes in, builds that, writes the code, runs it, tests it, but hits an error. It reads the error, it fixes it, and runs again. Like, it's self annealing, or it'll iterate until the thing actually works.
56:47And then you review it, tell it what to adjust, and adjust it. So it's like working with a developer except they can finish this in minutes instead of weeks, and it costs exponentially a fraction of what a developer costs.
57:00So the key thing that I need you to understand is that you don't need to be technical to use Cloud Code. You just need to understand any language in the world, and you need to be clear about what your business needs. Like, that's it.
57:11If you're gonna explain a process to a new employee, you're gonna explain it to ClaudeCode. It's the same skill. Now
57:18ClaudeCode, it runs in a terminal. For some of you, that sounds scary, and it's actually not.
57:22It's just a text window where you type things in and Claude responds. Like, that's it. And it connects to your tools.
57:28And you can use an IDE where an IDE is essentially just a magic interface where it's easy to navigate everything, type in plain English,
57:37puts everything out for you, handles everything, just makes it very easy to utilize everything. Anyways, there's two types of systems before we actually get into the build.
57:46I just need to make this distinction real quick because it'll change about how you think about what's possible. So two types of AI systems. Type one, workflow automation.
57:53So when x happens, you do y, then z. Like, that's a recipe.
57:58The steps are clearly defined. AI follows them exactly every single time, and there's not much thinking required. And it's just gonna execute.
58:06So this is what we're building first, which is a lead response system. So it's predictable, it's reliable, rents $24.07 without anyone having to touch it. Type two is AgenTic AI.
58:16So instead of just following a fixed recipe, an agent just looks at the situation, decides what to do. So you give an objective, not any sort of, like, checklist,
58:25and the difference is just like giving a task to an intern with a very detailed SOP as opposed to just giving it to an experienced employee who knows the business.
58:34Where the intern, they follow the steps, the employee, they are reading the room and figuring out the best approach. So for example, like a workflow sends every lead the same exact follow-up sequence. An agent, it can read the lead's website, look at what they said submitted,
58:48understand what kind of business they run, and write a genuinely personalized message. It's the same goal, but just with way more intelligence. So we're gonna build one of each.
58:55We'll do the workflow first, and then the agent. Alright. So now it's time to actually build these automations and go through these demos, but I'm gonna be showing you guys two simple builds that you can see the whole thing end to end.
59:05And if you guys want the deep technical breakdowns on how Cloud Code actually works under the hood, how to set up MCP connections step by step, advanced prompting. I've made all of those videos covering all of that on this channel, so make sure to check those out if you guys want a deeper dive into these demos, into those technical walk throughs.
59:22For now, just follow along. You'll see how straightforward this actually is to utilize. So the first thing I'm going to do is I'm going to utilize cursor, which is going to be my IDE of choice.
59:30You can use Versus Code. You can use Windsurf, whatever you want, anti gravity. And this is just going to be how I'm going to be interacting with Cloud Code.
59:38So I'm going to open up a brand new project. I'm going open projects, and I'm just going to create a new folder inside of our AI folder. I'm going to call this specifically
59:46we'll just say maybe leads response or lead response.
59:51Open that up, and here we go. We can close out all of Cursor's native features and everything.
59:57We just need to focus on what we actually need, which is going to be Cloud Code. So inside of here, to actually start installing it, you can just go to your extensions, you can find Claude Code, and you can install.
60:07Now from here, it just wants to log in through your subscription, or you can provide it with your API key once you actually open up a Claude Code. Now from here, all we have to do is we could just go to these three dots in the top right. We can open up the Claude code terminal, and I'm going to say, yes, I trust this.
60:22Now it's out of here. Claude code. It is just utilizing my terminal at the bottom, but, you know, we just have this nice little interface.
60:29But if he needs to close anything out or override anything, you could do so from this terminal down at the bottom if need be. Alright. So anyways, what we're going to be building.
60:38Let's imagine that you've got a lead intake form on your website. You know, somebody fills out their name, they give out their email, their company information, and what they actually need help with. So whenever they are clicking submit, within seconds, the system ideally is going to be checking if they're a good fit.
60:53It'll generate a personalized email for your review, for your team to review, and notifies all your sales team on Slack from there. So I'm going to have Cloud Code just build the whole thing out for me.
61:05So it's going to include the form itself, so we can test the full flow right here on screen. And in the real world, like, you would just connect this to whatever you already have on your website, whether that is Type form or a Tally or Go High Level form, Google Form, which is what we're going to be using for this. So for this demo, we're just gonna be building it all in one place so you can see the complete pipeline.
61:24Now before I start building, what I always do is I plan first. So we're not just going to be jumping in blindly. And if you're unclear on what you actually want, Claude code is going to be very unclear.
61:34So garbage in, garbage out. I always say this. So let me just go ahead and tell Claude code what I need and have it plan the approach before it's going to write any code whatsoever.
61:43So what I'm going to say is I want some plan and then build a lead response automation. Let me give you the full picture. Business context, we're a renovation company doing about $3,000,000 a year with 12 employees.
61:53We use GHL as our CRM, Slack for team communication, and Gmail for email. So here's what I need. There'll be a complete local web app that does this.
62:00So one, a simple lead intake form that runs in the browser. It fields for the full name, email, company name, number of employees, annual revenue, and what we are looking for help with. Make it look clean and professional.
62:12And when somebody submits the form, the back end process of the lead immediately checks if they match your ICP. And if they score a six or above, then draft a personalized email that references what they actually wrote in their form, and then you can send a Slack message to our sales channel. After the form submits, show a confirmation on the page with the lead's ICP score and what actions were taken.
62:33Use node. J s for the back end. Use the Gmail, Slack MCP connections that are already set up.
62:39Use the Cloud API for ICP scoring, and email drafting in one column. Before you build anything, walk me through the plan. So you could see, Cloud is automatically entering plan mode.
62:47But we could also just do you know, we could just type out something like plan inside of the terminal, or you put it inside of here and make sure we're going into plan mode. Also, if you do shift tab, you can cycle between plan mode and anything else.
63:01Okay. So right here, it's asking for some permission. So it's asking the Gmail and Slack tools are available to me in Cloud Code, but cannot call or be called directly by a standalone Node server.
63:12Also, the app handle Gmail drafts and Slack messages. Do we have a couple options? It's recommending the former, which is going to be Cloud Code in the loop, in which case the Node app is going to be handling the form and the cloud API scoring and drafting, and then returns the results to the browser and write to leave file.
63:26And then we can just tell it from there to process it, and I can use or it can use the MCP tools to create the Gmail draft and then send the Slack message. So we also have a fully self contained app that's looking like we're going to have to utilize something like Docker, which I do not want to go about. Then we have a hybrid.
63:42So we're just going to go with ClaudeCode is recommending, which is going to be number one. Okay. So here's what we just got back.
63:47It looks like this is the entire plan. We can see up at the top lead response automation implementation plan. Obviously, everything is very in-depth and built out pretty thoroughly
63:57as it appears. Now we have the architecture. We have a two step flow.
64:02So the first step is automated. This is the form submission going into the Cloud API where it's going to score and then draft things. Next up, we have the browser showing the confirmation.
64:10Then we have Cloud Code. This is where the user has to trigger Cloud Code to execute any Gmail drafts the Slack notification. That's where we're going to be utilizing MCP.
64:19Over here's some of the implementation details. So a little bit of information about the front end specifically, about the back end, the Cloud API call, the logger, lead files, MCP execution,
64:30dependencies. So I'm going to go ahead and give it permission. I'm going to say auto accept and move on to the next step.
64:37But from here, it's going to approve the plan. It's going to recognize that it needs to start building it. You may have to oversee this for a second to approve of any request that it is making.
64:46But in any case, it's going to start building out the entire front end, start building out the back end. Now it is worth mentioning, right in the side of this plan, this is where you would be catching any mistakes. So if nothing is aligning with your actual plan and how you anticipate
65:00going about this entire process and really just making sure that this is going to be ups to your standard, then this is where you would notice it. So at times, you do have to go through it thoroughly, or maybe you can even provide this to another cloud system, another cloud agent.
65:15I mean, it can be as simple as actually opening up another Claude code terminal, and just asking it to review the plan and making sure that it aligns with what you would what you want.
65:24So, of course, just provide it with what you want, compare that with the plan that it actually came back with, Make sure everything is sound and actually aligned with what you need. Proceed with this where it's going to install NPM. But anyways, former, it looks like it was building out the back end.
65:39Okay. So everything is now built out. We can see here's the specific files that have been created.
65:44Here's how we can actually start running this. Here's how the flow works. So to break that down pretty high level, one, when a lead is submitted, Claude is going to score them one through 10 and draft an email if the score is above and a six or equal to six.
65:57Two, the results show in the browser instantly. Three, the lead data is going to be saved to a leads folder as adjacent with the status as pending, and we just have to tell them to process the pending leads, and we'll go ahead and create the Gmail draft and send the Slack notification through an MCP server. So with that section right there, number four specifically,
66:14that is what we refer to as the human in the loop feature. Now we just want to add an API key. So what I've done is I just went inside of OpenRouter, which is just
66:25a magical place where you can get a bunch of different models so you get access to pretty much everything that is out there. So you have Anthropic, you have Open EIs, ChatGPT,
66:35you have DeepSeek. I mean, you have if you can think of it, you can name it. If there's an available model, it's going to be on there.
66:40So what I'm just gonna do is I'm just gonna throw in my API key and see where that takes us. So the thing about Cloud Code is it's going to recognize that that's not an anthropic key. So recognize that it's open router, so I'm just going to go ahead and say, yes.
66:53This is open router. That's fine. We can switch to using this or run that off.
66:57So in the meantime, we're going to open up localhost and see how everything is running. So it looks like the site cannot be reached, so that is, of course, very problematic.
67:06So we're going to have to let this know that the localhost, it's not up and running properly. So what I'm gonna say is the local host that you gave me is not running properly.
67:15I need you to ensure that this is resolved. So we're going to run that as well.
67:18We'll just queue that up whilst this is finishing. Okay. So one of the biggest issues is I didn't even have credits inside of my open router account, so I just threw in a few bucks.
67:26And now we have the server up and running. Now the reason that we're using localhost is just for sake of the demo to have something up and running very quickly. But of course, you guys can be using any other form connector, whether you want us to use Typeform, Tally, you'll be able to connect through that as well.
67:39But if we click on this logo host, we can see we have our form submission now. So we have the full name, the email, company name, number of employees, the annual revenue, and what they're looking for help with. So instead of here, I'm just going to submit some dummy data.
67:51So I'm just going to input Elon Musk. We'll do elan@x.com.
67:59Just call it x. Number of employees, we have over a 100. Annual revenue, I would say over 20,000,000.
68:06And what they're looking for help with, I'm just gonna say something simple. We need help automating our project management and client communication. We're growing fast, and our current manual process is breaking down.
68:15So let's submit that. You'll see that it was going to analyze the lead from there. Now they have the information.
68:21So here's a summary of how your submission was processed. We would actually probably not want to show this to the person actually submitting this form, so we can definitely tweak that a little bit, but it says that X is a tech and social media company, not a service based business.
68:38Revenue is above our sweet spot range. Employee count exceeds our target range of five to 50 employees. While they have a clear need for project management automation, they're not our ideal customer profile.
68:48So the action's taken. Slack notification is now queued. So what we're actually going to do is we're going to submit another lead and now have this one be actually qualified.
68:56So what I'm going to say is we'll do Nabin, and we'll just give it an email. Okay.
69:03And the number of employees, we'll do about 11 to 25. Annual revenue, we could do about
69:10anywhere between 1 and 3,000,000. And once again, we're just going to give it the same exact about message of, you know, what we're actually looking for help with.
69:19Let's see if this one's going to be a little bit more qualified. So right now, it's going to analyze the lead, analyze the specific submission, all the filters, and now it's recognizing that this is falling within our
69:32our qualification of an ideal ICP. So a strong fit, the renovation service business with about 11 to 25 employees, and 1 to $3,000,000
69:41revenue plus clear urgency around project management and communication automation needs. So actions taken, Gmail draft queued, Slack notification queued. Let's check that out now.
69:49Right off the bat, before we even check Gmail or anything, I'm going to go ahead and open up our leads folder, and see here we have two different leads. So if we open up Elon Musk, we should be able to find that it determined yeah.
70:02So the score, this was a two. So, of course, it's not going to send off the email draft or anything else. And for this other one, so Namine, for example, inside of here, status is pending, and the score is eight.
70:16So I want to ensure first that everything is connected properly. I'm going to double check and make sure all my connectors are hooked up.
70:24How I'm going to do that is I'm going to open up my MCP servers. So we just do slash MCP. You can see all the current ones that I have connected or some that need authentication and just reconnected.
70:37So we see Gmail right here. It is connected. And Slack, if we
70:42scroll maybe I'm missing it. Okay. So it looks like Slack is also connected.
70:46I'm just going to do a quick check, and it doesn't look like anything was actually drafted for the lead. So that's obviously problematic. So this is where we're going to utilize Claude to make sure that it's going to solve everything for us, and pretty much a self anneal.
71:01So I'm going to say, okay, everything is connected, all the connectors, but a recent lead that had just submitted the form, he was qualified, but nothing was actually drafted, and it doesn't look like anything was sent in Slack either.
71:17So I need you to double check that the Slack channel is actually connected properly. So I believe I have a sales updates Slack channel that you can interact and post things inside of there, and make sure you are drafting the email properly. So, of course, we're using some speech dictation there, just allowing me to effectively code and get things done even quicker.
71:38So it's going to check the pending leads, find the right Slack channel, then we're going to allow it to actually search through everything. Okay. Awesome.
71:45So it says both leads have been processed. Here is what happened. Nabin, he was scored.
71:50He was qualified. It looks like Gmail created the draft. We can also view it through this command right here.
71:54Actually let's open that up right now. Awesome. So it says, hi, Namine.
71:57Thanks for reaching out. Can we understand the challenge you're facing when you're going fast? Those manual processes that worked at smaller scale can really start breaking down.
72:04Automating project management and client communication, exactly the kind of challenge we love helping renovation companies tackle. And I'm just gonna skip to the end. I'd love to learn more about your specific workflow challenges and what you've tried so far.
72:13Are you open to a quick fifteen to twenty minute call this week to explore how we might be able to help streamline things for your renovation company? So that's looking perfect. Exactly what we're looking for.
72:22Now it's real quick, check on Slack. As you can see inside of our sales updates channel, we have a new lead, Naveen, and we also have new lead, Elon Musk.
72:30So now I'm sure some of you are probably thinking about the obvious or kind of the next step from here now that we actually have this. So everything currently, it's all manual. So when you close Versus Code, this thing is going to stop running.
72:44It's not actually twenty four seven. So what we built, this is just running locally. So if my machine goes to sleep, if I close this out, the server, it dies.
72:53And that, at the end the day, that's just a simple demo. Like, it's not really providing much value inherently. So we need to make this real.
73:00We need to push this into production. We need to deploy this so that it's twenty four seven inside of a cloud with some public URL that your your actual website form can be hitting. So I'm going to do it without leaving clogged code.
73:11So there's a lot of different options out there you can use, render, but we're going to be using Railway. This is just a cloud hosting service, so you can think of this as where your code is going to be living and running permanently without you having to manage the server. And what makes this perfect for what we're doing is Railway, they actually have an MCP server, which means that Cloud Code can just talk to it directly, and we don't have to really do much except for just make sure we have the account signed up for, have some credits added into the account.
73:36So I can just tell Claude to deploy this project, and it will handle everything for us. So let me set this up. First, I'm just going to add the Railway MCP connection.
73:43So I'm just gonna type out is add the Railway MCP service so we can deploy this project, then deploy this lead response system to Railway. It wants a public URL that will accept form submissions twenty four seven, and we'll run this off. Sweet.
73:54So Railway is now installed, and it actually looks like they have their own CLI now, which CLI, if you're not familiar, this is just a more revamped version of MCP where it's more efficient. Now we'll get into that a little bit later, but we now want to open up our terminal. We're just going to copy this,
74:10and we're going to run this off. So we're going to open up the browser and just do whatever it's asking. So we'll continue with our GitHub account, log in, make sure we're just providing it with the necessary credentials.
74:20We'll click allow workspaces authentication successful. So we should be good to go if we go back in to here.
74:27So I'm just going to go ahead and say proceed. Looks like it was unauthorized.
74:33So maybe we need to check on that, but let's see what's going to happen if we just type out proceed. Okay. So it is running unauthorized for some reason.
74:41So let's try that once more. We'll throw this in once again. We'll open up the browser.
74:45We'll authorize and see if this works properly. Alright.
74:49So we did get another error. So we're going to just try going browser list and see if we can activate it through here. So we're just going to open up this link.
74:58We just press command and click on it, and we have to enter our device codes. Let me throw that in real quick. Okay.
75:04So that did end up working. And if you run into any issues like that or this goes for really any other thing that's kinda menial, you just can't figure it out, that's where you're using AIs to help you troubleshoot and anneal any problems that you're running into. So we're just going to now
75:17type out proceed and see where this can take us. Okay. So everything is up and running.
75:22It looks like it built everything out for us. Went through quite a long list of errors and issues, but, you know, of course, since we're using Cloud Code, it's able and possesses the ability to self anneal, in which case it's going to solve for all these issues. So now it's going to accept all these form submissions twenty four seven.
75:42Sarah Johnson, she just scored a nine out of 10 with an email draft queued that it was testing itself, which is awesome as is. And to process the Gmail drafts and Slack notifications for leads submitted via the public URL, Just tell me to process the pending leads anytime, and it'll be able to send that off. So right now, everything's going to be working and running.
76:02So the only thing that I would wanna change from here is changing the form. So right now, it's just connected to a local host.
76:09We'll then wanna point that form to connect it to something like our website or Tally type form, go high level, whatever. But anyways, that is quite literally how simple it is to deploy and put something into production where you don't have to push it to n eight n do anything like that. You could just simply utilize an MCP server or any CLI to actually get things up and running.
76:27So that's the full loop. We now have this full form submission process, and that took maybe twenty minutes.
76:33Now what we're going to be doing is building something completely different. We're gonna have the same approach, but just a different type of system. By way, if you guys are looking for more advanced breakdowns and techniques
76:42and different strategies and really just looking for some sort of community, make sure to check out our free school community link. It's down below in the description. Anyways, let's move on.
76:50So what we're going to be doing this time, it's going to be a little bit different, but for now, I'm going to close this out, and we are going to create a brand new instance. So we're just gonna do the same exact thing as we did before. We're gonna create a new window.
77:01And by the way, you could do all of this through your terminal as well. If you want us to utilize Cloud Code through here, it's going to be as simple as just, you know, creating a new folder inside of here, then you just launch Cloud Code. Much the same exact way that you did it through cursor, except through cursor, you're just pressing buttons and choosing the extension and installing it through there.
77:19So it's just, you know, not having to implement any any commands, whereas inside of the terminal, you do have to utilize some commands. But anyways, we're gonna create a new folder, and we're going to call this the inbox agent. From there, we'll just open it up, and we'll launch Cloud Code from here.
77:34So because I already have Cloud Code installed, if you go to your extensions, obviously, you can see I have it installed here. You actually open it up. There's a shortcut instead of having to go to three dots up in the top.
77:44You can just do command escape, and this will also open up the Cloud Code terminal. Now it is worth noting that you do not have to utilize the Cloud Code terminal.
77:53So here's another option. If you wanted to, you can open it up. We'll go back into the settings,
77:59and we will go to scroll down to the bottom. You can deactivate where it's not going to launch in the terminal,
78:06whereas it's just going to be a little bit more UX friendly and better for beginners, but I prefer using the terminal. Anyways, let's back up and get into the second build. So this is gonna be just a little bit different from what we had just built out.
78:19So the lead response, that first demo, that was a recipe. So it has the same steps every time. It has the same order every single time.
78:27Now what we're building now, this is going to be an agent. So this is going to be a Gentic. So it'll look at each email inside of my inbox, and it'll make it a different decision about, you know, what it needs to do based on what it is reading.
78:40So naturally in any business landscape, a client email, it gets handled differently from a newsletter, which gets handled differently from an emergency. So it's the same thing where we plan first and then we're going to be building. So I'm going to be pasting this in.
78:53I want to plan and then build an inbox agent. Here's the context. Nabin renovation company, we use Gmail on Slack.
78:59Here's the goal. I want a system that connects to my Gmail inbox using our MCP connection, pulls my recent on Reds, and for each one decides what to actually do with it. Now if it is urgent, like an emergency or a time sensitive client issue, flag it.
79:11If it's a client email or a new lead, give me a one line summary of what they need. If it is a vendor or partner email, draft a short polite thanks. I'll review this.
79:18Reply and save it as a Gmail draft using MCP. If it's a newsletter or marketing email, just skip it entirely. And after going through everything, send me one Slack message using our MCP connection with a clean digest.
79:29So nothing's going to be sending automatically. Drafts only, I review everything. So this is just to ensure that it's not actually going to responding to any of my emails that are rather high priority or anything else.
79:40So what's going to happen next is Cloud, of course, is going to just ask a couple of clarifying questions, same thing as before, And we're just going to pick the recommended option on each tab and hit submit answers, and then it should showcase the plan from there. So right here, select target. Which channel should the digest to be sent to?
79:58For the time being, I'm just going to say DM to myself. The trigger, how should this agent be triggered? And it could just be, you know, like a schedule, maybe every single morning that this can run, and what counts as urgent for your renovation business.
80:10So I will do keywords plus known VIPs, and we'll submit all these answers. So we're just providing it with as much information, so it's actually actually getting smart enough to realize that it needs some more context.
80:22It needs some more direction to actually create the best version of this. Alright. So it has written up the plan.
80:27It is now ready to execute. How would we like to proceed? Let's review this really quick.
80:32So if we scroll up a little bit further, we have the context we provided it. We have the architecture where it's going to use a Claude code skill plus a scheduled trigger. Now why it's using a skill and not a script?
80:43It's because Gmail and Slack MCP connections are already live. The skill just uses them. Claude handles the classification logic natively.
80:49There's zero dependencies, it's easy to tweak. So here's some files to actually be creating, what it's going to be doing once it actually starts off, assuming that we approve everything. Here's the step by step flow.
81:00Fetch unread emails, read each email, classify each into one of the four buckets. So we have urgent, client or lead, vendor or partner, newsletter or marketing, and draft vendor replies.
81:10It's in the Slack digest. So let's see an example. So we have urgent,
81:15clients and leads. We have vendor drafts created. We have the skipped and the review for drafts.
81:21It's going to go ahead and start off with a Claudette MD. So Claudette MD, again, that's the brain. So this will contain VIP contacts, vendor domains, keyword lists,
81:31and I can edit this at any time. Schedule trigger, use Cloud Code's scheduled trigger to set up daily trigger. Weekdays at 7AM.
81:39Sounds about right. That should be fine. Action.
81:42We can also run the slash inbox scale, and the delivery results arrive as a Slack DM. Alright. So let's go ahead and run this off.
81:49I'm going to auto accept these edits, so it just keeps on printing out. But, you know, it is worth noting that sometimes things can just keep churning out, giving you wrong output. So that is why you may want to review each output that it is coming back with periodically.
82:01Okay. So this just finished running, and it looks like between the last project and this new one, Slack somehow got disconnected. So let's go ahead and sign that up real quick, make sure everything is properly connected.
82:15So it's going to link us to inside of our cloud platform. But, I mean, it looks like everything is good to go, so let's just do a quick check. Let's go inside of here.
82:24We'll do MCP. Looks like Slack is actually connected, so I don't believe there's anything we need to do here.
82:30I'm just gonna give it a simple input saying, it looks like Slack is connected on every front, and we are good to go.
82:37Everything just finished running, and we can see what was created. We have the claudet m d. We have the slash inbox skill.
82:42We have the schedule trigger. So if we actually explore these a little bit further, we could go into our skills. We have the inbox dot m d.
82:48So this is what is considered as YAML. Now something that we will be exploring very soon, which is claud's new managed agents, new depending on when you're actually watching this video. But this is what YAML looks like, this particular structure.
83:01But at the end of the day, it's just a markdown file. So it's just explaining up at the top, here's some context. So the name, description,
83:08here's more or less the prompt and what it needs to be doing. So it's giving it all these different sorts of information. So what this inbox skill is
83:16consisting of. And if you wanna create more skills, it's just as simple as using the skill creator. This is a pretty much, this is a template that Claude or Anthropic provides to you to create new skills.
83:27Anyways, if you look at our Claude MD, this is what it created for us. So, this is just markdown. This is just the the naming up here, some descriptions, how this works, configuration,
83:38and some additional context of anything that is deemed important and relevant to the outcome of any skill or any reference file that needs to pull from that. Anyways, we noticed that we have an issue with Slack. I'm going to, once again, open this up.
83:52I'm just going to uninstall this from the cloud, or we'll actually wanna go into this one. We'll open this up in the cloud, uninstall Slack, and reconfigure it just to make sure everything is actually good to go.
84:04So we'll disconnect this, and we're going to, once again, configure this. So it'll just prompt us to log in to Slack again. We'll click allow.
84:12And on the top right, you can see we are now connected to Slack, and we should be good to go. So anyways, saying that we can manage our trigger here, so I am curious what this would actually look like inside of here.
84:22Okay. So this is just going to link us to Claude Code's scheduled feature. So we have the Nabin inbox triage.
84:30So here are the instructions, exactly what it had written out for us, and the associated connectors. So if we click on edit, we'll actually be able to associate
84:39some more connectors. So if we go to Slack, we can add anything else that we want. If we wanted to integrate with Fireflies, Google Calendar, where it's gonna look at any of our meetings, anything like that, it can definitely do so.
84:49So we'll save this, and we will back out of this, and let's start testing this out. So I've got some emails here. A majority of these are just sponsorship emails.
84:56A couple of them are client engagements, and then some of them are just internal things going on inside of our agency. Now what I would like to do is I would like to, you know, go through some of these emails, and just ask Claude Code to go through them and associate and assign them, and determine what it should actually do with them, how it should be responding.
85:14So what it'll look like is I just came back from the schedule. So instead of here, I want to get a little bit more specific. So what I'm gonna say is
85:21I want you to check my email inbox, use our skill that we had just created, pull my last, let's say, 16 emails, categorize each one, and send me the digest on Slack just like we had planned.
85:32So we'll run this off. Alright. So everything just finished running.
85:35Let's take a look at what this actually provided back to us. So it says, good. I've read all the key emails.
85:40Let me classify and send the digest. Here's the classification of the 16 emails, and let's just open up Slack so we were able to get it connected properly. Inside of here, here's our inbox digest.
85:49So the urgent ones, something with a sponsorship, and then we have a couple of clients and leads. So some people sharing some files with me, follow-up on a brand deal inquiry, and then it skipped about 10 newsletters and automated emails.
86:02And then there's a few ClickUp notifications about some things, and that is our inbox. So at the very bottom, you'll see that there is no vendor drafts created this run.
86:10Three emails were sent or draft or already read and excluded. So in this specific instance, there wasn't anything that it needed to respond to or anything like that. But it'll be just the same exact case as the last demo if we or if it recognized that it did need a draft, it would respond to them.
86:24So in that case, it was if they were, you know, included inside of our ICP, if they qualified, then it would create the draft for them. In this case, it deemed that there wasn't anything to send off, but it would work the same exact way. And we could even take this a step further if you wanted to actually categorize everything inside of our email inbox.
86:40I mean, it's just going to be as simple as providing the instructions to Claude code. So nine times out of 10, you can find out if something's going to be possible to do just through talking with Claude Code. And nine times out of 10, it's going to be able to actually do that task for you.
86:54And if not, you know, there's always going to be some sort of workarounds where it's going to be a little bit automated. But in any case, you just have to utilize AI as a weapon for what it actually is. But in any case, you just have to look at AI as a weapon because that is quite literally what it is.
87:08So you guys just saw me tell Cloudco to effectively go check my inbox, and it did the entire thing for me. Categorized every email, drafted replies, sent me a Slack digest.
87:18Like, that's incredibly powerful, but I had to tell it to actually go and do that. I typed the command, and if I don't type it, it's not going to run. And the whole point of this is that it should just show up inside of my Slack every single morning at 8AM without me having to think about all this stuff.
87:34So with build one, the lead response system, the deployment, pretty straightforward. Right? We pushed it to Railway, it got a public URL, The form is going to hit that URL,
87:44and it's going to make it run consistently and automatically. Now that works because the lead response system, it is reactive. So it is sitting there just waiting for somebody to actually, you know, knock on that door.
87:54In the inbox agent, it's a little bit different in the sense that nobody's going to be knocking. Nobody is going to be submitting that form. The agent needs to wake up on its own.
88:03It needs to go out and do things. It needs to check my email. It needs to make decisions.
88:08It needs to take different actions. And then from there, it needs to report back to me or report back to my team. So there's no real incoming request that's always going to be triggering it aside from that morning schedule.
88:20So the agent, it initiates the work, and that's a fundamentally different kind of system. And that's where something brand new is actually going to be coming in. Now that's where I'm going to be introducing Claude's new managed agent.
88:30So depending on when you're watching this video, as of when I'm recording this, this just released yesterday. So it's a brand new feature, and I wanted to include it within this course. I find it to be very important.
88:40So what Anthropic released in the simplest way that I can describe this is managed agents. It's just a platform that runs your AI agents for you inside of the cloud. So you can define what the agent actually does, what tools that it's going to be connecting to, what instructions that it needs to be following,
88:59all that good stuff. And in Thropic, it is going to handle everything else. So it will handle the server, it'll handle the security, it'll handle the credentials.
89:06So inside of here, you can see I have the credentials. Vault. I have the environments.
89:09It'll handle error recovery, the uptime, you know, all of it.
89:14So if we click on one of these sessions, you can see like a debug section, just like a full transcript of everything that's actually going on. So if you think about the difference of what we had done to this. So with Railway, that's where you're hosting a web server that's going to be waiting for any requests.
89:30Right? With managed agents, you are effectively running an AI agent that goes out and it's going to be doing the work for you.
89:36So, you know, compare this a little bit to real roles. Realway can be a receptionist that's sitting at a desk just waiting for the phone to be ringing.
89:45But managed agents, this is an employee who shows up every single morning. They're going to be checking what needs to be done, and it automatically does it. Right?
89:52So for this inbox agent that we had previously just created, that is exactly what we need. We define the agent, so you check Gmail every morning, categorize the emails, draft the replies for the vendors, and send a maybe a digest to Slack. We can connect it to Gmail in Slack, through the MCP,
90:09managed agents, it is then just going to configure all of that, and it's going to just figure all of it out. So every single day, you'll run it without us having to touch it. So there's no VPS, there's no Docker that we have to use,
90:22no Chrome jobs, there's no SSH ing into your server at midnight because something crashed and you have to investigate the log. It's none of that. Now with that, wanna break this down even further.
90:31So it's actually a lot simpler than it sounds. So there's actually four different concepts. That's all there is.
90:37But first, we have the agent. So what I actually have pulled up right here. So the agent, this is just the job description for your AI employee.
90:45So we can create a new agent. You can see, like, this is what it looks like. It's a YAML file, or you can also just have it be JSON.
90:52And we could also go to create our own agent where it's going to be a template. So they have blank agent, deep researcher, structured extractor, incident commander, field monitor.
91:01And if we click on any one of these, it's just going to have, you know, the YAML actually filled in for us with whatever is going to be relevant for this particular template. Right? So for support agent, it's gonna look a little bit different, be more curated towards what a support agent would require and what it needs to consist of.
91:17So maybe this will have Notion connected to it. And let's see. Is there anything else?
91:23No. It looks like that would be the only tool. But if we were to go to a different one,
91:27like this incident commander, it has Sentry. It has Linear.
91:32It has Slack. It has GitHub. So at the end of the day, this YAML is just consisting of all the context and all the servers, all the tools that it actually needs inside of this agent configuration.
91:42If we actually back up a little bit from here, we can just describe what we want the agents to be consisting of. So inside of this, like, we can say something like, you're an inbox management agent. You read my Gmail through MCP.
91:54You categorize every email. You draft vendor replies, and then you can send a Slack message or Slack digest to the sales updates channel or just DM me, and that's your agent definition. So I'm actually going to type that out explicitly how I just
92:07enunciated that. So we'll generate this. I'll show you guys what you can get.
92:10Alright. So this is what we get. It just created everything for me in the YAML format.
92:14And if we even open up JSON, I mean, it's gonna be the same exact thing. It's just gonna be in JSON format. So this one, it's just going to be utilizing Slack and pulling from that Slack MCP so we could create this agent as well.
92:27And boom, just like that, you could see Slack. And we have any built in tools that are going to go right here as well. Now next up, we have our environments.
92:34Now inside of an environment, all this is is just the workspace that your agent is going to be running in. So you can think of it like setting up a new employee's computer.
92:42So, okay, well, what software do they need? What access are they going to be needing more importantly? And, you know, the platform from there, it's going to spin up a very secure
92:51cloud container with everything pre installed. So it'll have Node. Js, it'll have Python, and whatever your agent ultimately needs,
93:00plus network access to all the tools that it talks through through MCP. Now third, this is going to be the session. So I briefly showed you this earlier.
93:08So we have the transcript, and we also have the debug. So this is just your agent actually doing the work. So once we spin up the agent, it's then going to create a session,
93:18and we can run it inside of here. So previously, I don't remember exactly what this one was. I believe it was an outreach agent where it was yeah.
93:26So we actually had an Airtable full of a few different leads. So I had three leads in there specifically.
93:32And this agent's role and job was to research the leads, look at the company information or any contacts who provided it, and try to create an output of a personalization line and email.
93:45So if we actually open this up, we can see here's what it did. So created some draft emails.
93:49So here's the personalization line, but better yet, here's the full email that it generated. Saw your go to market newsletter. It talks about the pain point and the solution that we do, and we ask we just have a low friction CTA.
94:03Like, is it worth thirty minutes to jump on a call or something like that? We did it for a few other people as well. So once the session actually starts, like, agent can wake up.
94:10It can hypothetically check your email if that's what you want it to do. It can process everything, send that digest if we're gonna relate this back to the inbox agent that we created in Claude code, and, you know, it'll be finished. So a session, I mean, could run for minutes, it can run even for hours, it really depends on the job that you're doing and what you assign to it.
94:27And you could also schedule these tasks to just kick off, like, whether that's every morning at 8AM, every hour, really whatever rhythm that your business is going to be needing. Now fourth, this is going to be actually inside of the sessions.
94:37This is going to be the events. So instead of the events, this is where and how you actually talk to a running agent or how you can be monitoring
94:46what it's actually doing. So you can send the message mid session, like, hey. I need you to check my sent folder, and it it is going to adjust.
94:53Or you could just watch the stream and see exactly what it's doing in real time. So it's out of here. You can actually check, like, sequentially what's going on.
95:01So here's, like, the very beginning, and here's the latest output. Now to take a step back because I wanna compare, like, what we built with inbox agent inside of Cloud Code with this,
95:11you know, managed agents or whatever they're calling it. So inside of this because, I mean, it's a little bit confusing, like, the discrepancy. Which one should I be using?
95:18Is managed agents better or what what process that we just went? Is that better? Now with all this being said, I know a lot of you guys are probably thinking, because this is exactly like what I thought previously too.
95:26It's like, okay, well, didn't we already just schedule the inbox agent earlier in Cloud Code? Like, how is managed agents different from those two demos that we had just done? And, yes, we did.
95:36Like, so the distinction between that is, like, what we set up earlier, it runs through Claude's built in scheduler. So it works great. It's fast to set up.
95:44And for your first automations, it is the easiest path. Managed agents, though, this is going to be the production grade version of the same idea. So the difference is actually, you know, what happens when things go wrong or when you scale things up.
95:57Where managed agents, it gives you very secure credential vaults, your API keys are not just gonna be sitting in a config file or anything like that. It's going to give you automatic error recovery, so if something fails at 3AM, like I mentioned earlier,
96:10it retries instead of silently just dying out. And it also gives you session tracking, so you can see exactly like what every agent did and when. So it's just built to handle dozens of agents running across your whole business.
96:22It's not just one or two. So think of it this way. The scheduler that we had just set up a few minutes ago, you know, that is just like setting a reminder on your phone.
96:31They manage agents. It's like hiring an operations manager who actually is able to handle everything for you in one place. It is going to report back to you, never drop the ball, or rarely able to drop the ball.
96:41So I would recommend to start with a scheduler. It's a little bit more UI friendly. And for scale, it's going to be best because everything is concise.
96:48It's all going to be in this place right here. So when your business is running ten, fifteen, 20 different agents and automations, you need some bulletproof infrastructure
96:57directly behind them. And that is where managed agents is going to be great for a majority of you guys watching this and your business environment. Alright.
97:04So now you guys have gotten the full picture, and I wanna make sure this is actually crystal clear because this is like one of the most practical things that you guys will be taking from this video. So again, there's two types of AI systems that you will build for your business, and each one's going to deploy a little bit differently for type a.
97:18That's going to be the reactive systems. So these sit and these wait for something to be happening. So if a form gets submitted, if a webhook is firing off, if a customer sends a message, you know, the system responds, and that's your lead response system from build one that we just showcased.
97:32So for these, you can deploy it to a hosting platform like Railway where it will get a public URL. It'll run twenty four seven, and it can process any incoming requests the moment that they will be arriving. And with this, can just deploy it straight from quad code with Railway's MCP server.
97:46It'll just be one prompt. It's really easy. And then type two as the proactive systems.
97:50So like these, they're not gonna be waiting for anything. They can wake up. They can go out.
97:53They can do work, they can report back to you. And then, you know, check the inbox. You can scan the CRM, generate a weekly PNL report.
97:59Like, that's effectively your inbox agent from build two that we showcased. And for these, like, you can use managed agents where you can define the agent, you can connect your tools through MCP,
98:09you can set a schedule, and in Thropic, it'll run it for you. And then reactive systems, they will need some sort of door.
98:16So railway, it is going to give them that door. And then proactive systems, it needs a brain that shows up to work, and that's where managed agents is going to give them that exactly.
98:25So between those two deployment paths, you you can take literally any automation that you build with Cloud Code and get it running into production in, you know, seconds just like that. And with all that being said, you now know how to build and deploy both types of systems. So let me show you what else is going to be possible because what we built today, it is just the starting point.
98:44So those two builds, they just showcase the method and how we actually go about it, where you describe what you want, plot code, it plans it, it builds it, you can test it. And the same approach, it works exactly the same way across your entire business. So let me give you a sense of what's actually possible, and I'm gonna give these pretty quick here because this could be over three hour video if I wanted it to be.
99:01So marketing. This is where AI can scan your industry every single morning and surface content ideas that are actually relevant. So for example, you can approve one idea, it'll automatically become social posts, maybe an email for your newsletter, maybe a video script outline, or your newsletter, it can write itself in your voice, you know, review things, you can hit send, and the SEO content pipeline.
99:20Like, this can run on autopilot with any keyword research. It can do draft generations and also a ninety day refresh cycle that catches any outdated content before Google does. Now for sales, this is where things can get even more powerful.
99:31It's where AI can build your prospect list, it can enrich it with company news, hiring signals, right, outreach emails that are going to reference something specific about their business, not something just like, I'd noticed that your company is is growing, but instead saying something more in-depth, like, I saw you just opened a location in Denver and are hiring three project managers.
99:48That usually means you're scaling delivery and the ops are getting stretched. Like, that's a completely different email, and it's going to perform so much better. And then from there, after every sales call, the AI can generate notes and update the CRM, send out proposals.
99:59We'll first create those proposals, then send it out, wait for any responses from those people. I mean, there's unlimited amounts of potential and opportunity with what you can do inside of sales, and not even just sales or marketing departments. It's really anything outside of that as well.
100:12So like customer service and finance, You can have AI handling about 70% of support tickets instantaneously. It's not with just any canned and templatized responses, but with actual answers pulled from real customer data.
100:24Can have a voice AI system that takes any after hours calls or any calls that you're missing, or just takes all the calls that you guys currently are taking right now and reach out to people. So this is something that we do in sync too, which is a separate holding of ours where we implement voice solutions and other front office solutions
100:40for health care clinics. And we tie that in with marketing where we run ads and we get them more leads, and then use AI to engage with all those leads, whether that's calling, texting, emailing. I mean, we interact on every single channel, and that's just going to show what you can actually do within
100:55all these different types of departments. So none of this is theoretical. None of this is science fiction.
101:00Every single one of these is buildable with the stack that we just covered. And some of them, they take an afternoon. Some of them, it might take a week or a month, but they're all real.
101:08They're all running businesses right now. Like, you look at these things, you might think that it is impossible or, you know, there's so much of a gap, but there's really not the technology. The you know, everything is just democratized to actually build up all these solutions.
101:21So that gap, it is just narrowing every single day, and like we're seeing with these constant new releases from mostly Anthropic how easy it is to build this stuff. And managed agents, that's just another example. Now ideally, once you already have, you know, five, ten, 15 automations running, you really should be having some centralized
101:37place that you can oversee everything. And this brings us full circle to what we just talked about in part one. That's the operator to owner shift.
101:44So this is where you have a command center, and that is going to operate as your dashboard. So you can use that as three different colors. Or green, you can run things automatically,
101:52or things are running automatically with no issues. And this is the stuff that you don't touch. It's just, you know, working for you.
101:58Yellow, so this is where AI is flagging something that needs your decision. Nothing urgent, but just to handle it within a couple hours. And then red, this is any immediate human action needed.
102:07So, you know, this can be your morning routine where you open the command center, you can handle the reds, there might be one or two, and then review the yellows, there might be a little bit more, five, maybe six, Then just scan the greens to make sure the numbers actually look right. And that could take you as little as thirty minutes a day as
102:21maybe it previously took you three hours, four hours a day. And I mean, those are exponential leaps in terms of your day to day. So for example, we're doing this for a current hospital where we're providing them with a financial dashboard.
102:33So just some sort of command center where they can see all the invoices, all the ones that they have to follow-up on, and really any other analytics inside of their hospital. And it's a pretty big operation that they're running, so they have more than a 100 people on board, and I mean, they're seeing thousands of people every single day.
102:48CTASo I mean, with something like that, with this scale, there's so many different systems that are flowing into this, and so many different things that they need to be looking at. So for them, I mean, they're spending over ten hours a day just, you know, having to figure out what is actually going on in this centralization.
103:01CTAIt's allowing them to actually not get blindsided by pretty much every piece of data every single month. So that is what it could look like for your team as well. And I just covered everything that you need to know to get started building that stuff.
103:12CTABut anyways, to kind of wrap things up, the thing that I really want you to hear is that this gap between businesses that are succeeding with AI and those that simply are not, it's not a technical skill. It's just whether or not you actually start. The tools, they are here.
103:26CTAThe process, it is very clear, and the only thing left is you actually doing it or finding some help and just, you know, taking that leap. So if if you guys do want us to keep going, and I hope that you do, we've got over 18,000 business owners all building this stuff together inside of our free Skull community. Link is down below in the description.
103:43CTALike, people here are sharing the builds, asking questions, helping each other, starting their own AI business. It's the best place to just keep the momentum going after this video.
103:50CTAAnd if you wanna stay plugged in beyond just this video, I break down AI implementation strategies, new tools, and real builds every single week in my newsletter, the AI course. Link is down below in the description.
104:01CTAIt's completely free. And if you guys got value from this, and I hope you did, because I put everything that I know into this one, like, subscribe, it helps more people find this video. And by the way, if you are a business owner and you're looking to have us just come in and implement all of this into your operation, then you can book in a call with our team to see how we can help you guys scale in 2026.
104:18CTASo link will be down below in the description. Alright. I'll leave it at that, and I'll see you guys
— full transcript
§ 05 · For Joe
The correct order for AI implementation is audit, data, tools, build.
WHAT TO LEARN
Most AI projects fail not because of the model or the tool but because teams start with the tool — the correct sequence runs the opposite direction, and reversing it changes everything.
- Audit your processes before buying anything: two days of live task logging reveals the invisible repetitive work consuming 50-60% of most workdays that never appears on any dashboard.
- Score automation candidates by multiplying frequency, time cost, and simplicity — the highest-scoring process is your first build, not the most exciting one.
- Automating a broken process produces broken results faster: document the process clearly enough that a new hire could follow it before you hand it to an AI.
- Data hygiene is not a one-time cleanup — three standing rules (standardize naming, mandatory fields, monthly audit) keep entropy from undoing the work within 60 days.
- The correct data layer depends on revenue scale: Google Sheets works up to $5M; relational structures like Airtable or Notion serve $5–20M; Supabase or Postgres for larger operations.
- The four layers of an AI stack have a strict order: single source of truth at the foundation, then LLMs as a swappable brain, then an agentic builder like Claude Code, then specialized tools added only when the backlog demands them.
- MCP eliminates the custom integration problem — one connection per tool means any AI agent can talk to any tool through one universal standard without glue code.
- Reactive automations (respond to triggers) deploy to a cloud host; proactive automations (wake up and do work on a schedule) deploy to a managed agent platform that handles uptime and error recovery.
- The business case for fast lead response is not intuitive: responding within one minute produces a 391% increase in conversion rates, while the average B2B company takes 47 hours.
- BCG research shows 70% of AI implementation success comes from people and process changes, not from picking the right algorithm — most teams do the inverse.
