§ 01 · The Hook
The bait, then the rug-pull.
Most productivity systems tell you what to do. This one does the research, drafts the emails, compares the insurance quotes, and finds the birthday gift — then hands you a tray of approvals. Built entirely inside Claude Code with MCP connectors and a handful of custom skills, the LifeOS dashboard turns a scattered inbox into a five-minute morning ritual.
Where the time goes.
00:00 – 00:21
01 · The morning-brief promise
Hook: all admin pre-handled, one clean list of approvals waiting at login.
00:21 – 01:30
02 · The 4-pod concept
Today, Money, Admin, Social — four lanes that cover everything outside the workplace.
01:30 – 03:35
03 · How LifeOS actually works
MCP connectors feed skills; skills feed the dashboard; dashboard pushes to Telegram.
03:35 – 08:46
04 · Dashboard walkthrough: Today tab
Email reminders converted to actions, action queue with Approve/Skip/Defer, 72-hour calendar view.
08:46 – 11:53
05 · Under the hood: CLAUDE.md and the personal-inbox skill
Project structure, context file, the triage skill that classifies every email and routes items by pod.
11:53 – 19:34
06 · Money, Admin, and Social pods
Lunch Money API for finances, document harvest for accounting, relationship-intel skill for birthdays and social events.
19:34 – 22:18
07 · The commodity argument
Gemini Spark and Anthropic Kairos are coming. The tool is already a commodity; domain expertise is the only durable edge.
Named ideas worth stealing.
00:21
model
The 4-Pod Model
- Today
- Money
- Admin
- Social
Maps all personal life admin to four non-overlapping functional lanes, each with its own data sources and skills.
Steal for any personal productivity system or client onboarding framework
01:30
model
Skills-to-Dashboard Architecture
- personal-inbox
- calendar-intel
- personal-finance
- document-vault
- relationship-intel
- action-queue
- session-prep
- notifications
Each of the 8 skills writes to exactly one section of the dashboard; separation of concerns prevents skill conflicts.
Steal for any agentic system where multiple sub-agents contribute to a single output surface
05:08
concept
Human-in-the-Loop Approval Model
Agent prepares and researches every action but never executes on money or communications autonomously. User sees Approve / Skip / Defer for every item.
Steal for any AI workflow that touches finances, communications, or irreversible actions
Lines you could clip.
05:08
"I want to have a human in the loop for all of this kind of stuff. I don't trust it with anything to do with money or handling my emails. You will never write a complete email on my behalf."
20:33
"The AI operating system as itself is a commodity. Anyone can build one. That is not where your value lies."
21:01
"The value comes in your experience within a specific domain, something that makes you more valuable than the person next to you who can just talk to Claude and get this thing to build anything."
Things they pointed at.
Word for word.
00:00Wouldn't it be great if all admin in your life just took care of itself so that by the time you log in in the morning, every bit of backlog, every area of your world is waiting for you in one clean list of approvals and actions? That's what Life OS does. In this video, I'm gonna walk you through how I went from scattered and overwhelmed digital life to building my own system that requires almost no effort, and I'll give you a prompt to get you most of the way there.
00:21Let's get into it. Okay. So before we get into the dashboard itself and start looking under the hood at how everything works, we need to understand the concept as a whole.
00:28And to do that, we just anchor it to pods. In this case, we are using four pods, but you can extend them depending on how much stuff you want in your LifeOS. The reason that I use this kind of pod mapping is because it helps us logically break down what skills we need or what specific actions are gonna be taken within the different lanes of our life or business.
00:45If you've been following this channel for a while, you know we've been building the AIOS model for businesses for a few months, and this is a very similar concept to that where we have the pod mapping but for business lanes. So I've taken the same concept here because in life there are very specific things that we deal with.
00:58Pod number one is all about today. So this is everything that only needs to be done today. Money is everything to do with money, your accounting, the projections, and things like that.
01:05Admin is literally any piece of admin that needs to be done, any documents that need to be filled out, stuff like that for any area of your life. And then social captures everything about your inner circle so you'll be able to remember everything without actually having to remember it. But when we think about these four pods, it's not just about having a dashboard that tells you things you already know.
01:23The goal here is to take action so that by the time this information is presented to you, we know everything we need to know and we can just do what needs to be done. So at a high level, it works very simply. You're going to set up a whole bunch of connectors via MCP and API,
01:35then you're gonna build some skills around those connectors and the things that we need to do for the different pods that we've mapped out, and it's then just gonna chuck it all into this dashboard that you've seen on the screen. After it's done that, you can then also get any daily message pushed to Telegram on your cell phone.
01:47You can approve and deny things on there, and also see what your day is going to look like from a life perspective. Like most things with AI at the moment, the only thing you really need to understand is how to break a complex problem down into natural language parts that you can then turn into an app or something like what we're about to build now.
02:02So we flip back across to my dashboard now and we're gonna start exploring this and then we'll dive into the back end and see how it's all connected and I'll show you the things that you need to connect as well. Something to note here, this is all mock data because I'm absolutely not going to be sharing my personal life, my financial life, and my family life with strangers on the Internet.
02:19So we'll be exploring my made up family over here and all of the other illegitimate things that are going on. But just know that it is a one to one clone of exactly what I would be doing in my actual life OS, which is in this little tab right next door over here. So when we open our dashboard, the obvious first thing we want to see is today because when I open this, I wanna know exactly what needs to be done and why I need to do it.
02:36So the concept for me came here, I spend so much time going through my inbox, but I also set myself reminders in my Gmail because I'm always checking my email.
02:45So it's really the only reliable place for me to figure out where the things are that I actually need to take care of. So when I sat down with that and realized that I had like 33 reminders that I never ever bothered checking, I thought, okay, we can build something that not only displays where we've gone wrong, but it gives us actionable advice before we've actually even done anything.
03:02So this top pane over here, what it's done is it's gone through all of the emails in my inbox. We have set personal reminders or where I've left things that are starting to become overdue. Depending on what the action is that I need to do, I can obviously click on these buttons.
03:14These you could change to whatever you wanted to. For me, I have drop it. So if I've already done this and just didn't delete an email or take care of something, I would do that.
03:21It would go and delete my email. If I click schedule, it would then schedule this for later by opening up my calendar, or I can click do it now and it would take me to my email or wherever the thing is that I actually need to take care of. That way, I'm taking action and getting to the place where I want to be.
03:33What's more important here and the value that I see throughout the rest of this LifeOS thing is that for the notes that I've said for myself or even for some of the action queue that I've got down here, the agent has already done everything for me and I just need to approve the thing. So it's not like I'm going out there and I just cloned whatever was in my email inbox.
03:50I've gotten the agent to take sentiment analysis, understand exactly what the context is of the message, and then in some points do research or take action and figure out the information that I need to achieve my goal by me either just clicking something or if I need to chatting further with it. So for instance, the first thing here, cancel the storage units, clear task, Life OS turned it into an action and checked the detail for you and your contract is month to month.
04:12So there's no notice period. It's in your queue below ready to approve. And so you can see that if I scroll down here, we'll have an entry in our action queue for canceling the storage unit.
04:21Then all I would have to do is click over here and I could click approve and it would take me to my imaginary storage unit site where I could sign in and cancel the contract that I had. Again, we're just showing off the fact here that instead of just having reminders, we want the action taken for us so that it is easier than ever to do the task.
04:38That's the whole point of this. This section up here might not be that valuable for you if you don't email yourself reminders like I do, but this section over here certainly is. This is the action queue, and what this is doing is it's going out to our Gmail inbox.
04:49Again, finding everything that is in there, and this would be from anyone who's emailed you, and then through sentiment analysis, figured out what the email is about, gotten the context, figured out how to match it to the context that we've given it about who we are and what we do, and then it knows exactly what to suggest for us.
05:02So with my home insurance renewal over here, I can literally just click on this and it would take me straight to my lemonade insurance where I could do whatever needs to get done instead of having to do it the old way. Now at this point, I realize you might be saying, well, why the hell didn't you just get the AI agent to do all of this for you so you don't even have to deal with it all?
05:18And you probably could. There are tons of ways you could get this done, but for me, I want to have a human in the loop for all of this kind of stuff. I don't trust it with anything to do with money or handling my emails.
05:26You will never write a complete email on my behalf. This thing will tailor drafts for me. So if someone's emailed me and I haven't gotten back to them, it'll automatically tailor a draft for me and just leave it in my drafts folder.
05:37Then in the same way that we have actions over here that need to be taking place, I can just click approve after it shows me a little snippet for the email that it's written. So for me, this action queue is probably the most valuable part of this today panel because this is gonna help me get through all of my backlog for this specific day so that I don't end up with a ton of reminders.
05:55Then on the other side over here, you can see what we have going on in the next seventy two hours. I think this is really valuable so that you understand what your week is gonna look like three days ahead. More importantly, there are actions that we can take on this when we get into some of the other menus.
06:06Then down here, it will give you a little view on what your money is looking like this week. Again, we have deeper dashboards into this. Same thing over here for social.
06:14This is everything we have coming up this week with people, and again, we will have some magentic work that takes place with each of these events. But before we get into the other tabs, let's just take a quick look at how this all works under the hood. First things first, we have our clore.md,
06:26and you can see it's just a little bit of information around how this project works and what we're actually doing here. So we've explained this is a personal life operating system, documents, finances, whatever that this thing is going to be dealing with.
06:37We give it a project structure, so we tell it that the actions are stored in a specific file. Use these to take action. It's the state management that the system is using.
06:46We don't even need to use a database for this specific part. Then we have context, and this is obviously giving it the information about who we are, what we do every day, who our family is, information about extended family members, things like that because it feeds all of the different tabs inside our dashboard. We then have data, which is just local SQL like databases that store specific things to do with finance and payments,
07:06any documents that we might need stored locally for specific reasons, things like that. We then have memory as well. So if this thing needed to capture any specific memories while I'm working in here, it would do that and use it for context later on if we needed it to.
07:18And then just the standard deliverables and temp, which is just exactly what it sounds like. Anything the system delivers gets stashed in there. But these are used less often because this is more of an action based dashboard as opposed to an AI operating system like we normally talk about on this channel.
07:31So without going through this whole file, all we're doing here is basically telling it how I want it to behave, how their live operating system actually works, where everything lives. That's pretty much it in a nutshell. Where the magic really starts is with each of the skills that map to each of the pods.
07:44So our first pod, which is the today pod, is actually a really simple problem to solve. It ties in with the admin pod, so I'm just gonna talk about them interchangeably because admin obviously forms part of the today pod.
07:54But really, all you need is a email inbox, and then all of your external systems send everything to Gmail because that is where we can pull our invoices, that's where we can pull any email or any information that someone has once sent us. So it becomes the central repo that we connect to with with MCP. Once that is done, all we need to do is have a skill that analyzes everything inside our Gmail inbox and then sorts it according to what we actually want done.
08:17The same thing applies for our calendar. This also gets connected to via MCP. Once we have that, we can build a calendar intel app and that also ties in with our personal inbox skill because of the various aspect and people emailing us saying, hey, can we do x y zed on Monday?
08:30The calendar intel app will piggyback off some of those emails and automatically set us up a calendar event as part of our social pod. And then finally, we have the action queue and this is the part where the agent analyzes all of this stuff and actually takes action for us like I've been speaking about throughout this entire video.
08:45So in terms of MCP, there are two ways that you can connect it for your life operating system. The first one is the official way using connectors over here. You can see I've got Gmail, Google Calendar, and Google Drive.
08:54Easy as done. You can also set permissions in here, which you absolutely should be doing. By default, you can't do any form of rights here, and that's what we want.
09:01We don't want this thing to be able to ever send an email, at least from my perspective. So you would need to go through this and decide what you need or Claude can help you do that while you're building this thing. Your other option, if you want super control over what you're doing to literally do whatever you want to do inside Google Workspace,
09:15would be to grab this MCP server. I will drop it in the description below. It's pretty reputable.
09:19I've been using it for a while now on the back end because I don't always rely on the official MCP servers if I'm testing new things. So this is a really good way to use it if you want to be a more power user. But just note, you absolutely don't need to use this for what I'm showing you in this video.
09:31I'm not gonna go through every single one of these skills because we'll be here too long, but I'll show you the one that does the personal inbox. So all we're doing here is we're telling it that it's going to triage the email in one sweep. While it's doing that, it's going to categorize, analyze sentiment, flag deadlines,
09:44and route items including attachments to the right life OS skill. And that's very important because like I said, different skills do different things. So we're telling it what it handles.
09:52We're telling it what it can do with Gmail, MCP, and then we're giving it a very clear definition of what the workflow is. First, it needs to search, then it needs to read. They get classified.
10:01Once they get put into a specific category, we can then analyze what's inside them. After we've done that, we can then send it off to another skill so that that skill can go out there and do what it needs to do.
10:10Most importantly on this one is we have a triage report because not everything is weighted equally. So we have specific things about when things need to take place. Of course, if there is a bill that really needs to be paid, that would be pushed up to the front of our dashboard.
10:23Same thing for anything that seemed urgent and is really old and I haven't touched, that would pop up as urgent, and we need to take care of that in the today part of our pod. And then through various scripts and m c p on the back end and the front end, we are routing things to different parts of the system until ultimately it all ends up in the part of the dashboard that it needs to be in.
10:40But everything stems from the Gmail inbox. And before I jump on over to the money tab, I just want to talk about the admin tab for second. It's mostly very similar to the today tab.
10:50It's just more generalized. It's not urgent stuff that only needs to happen today. But it still has the same views for things that we need to take care of, just not that urgent, and it will also create actions when those things need to be done.
11:01But there is something else that this thing serves as well, and that is to harvest all of our documents from our Gmail inbox. That's super important when it comes down to accounting because otherwise you would have to go and manually search for all of those things.
11:12For instance, if you have your own business, you would know that you get to claim a whole bunch of working from home expenses if you use your home as an office. So for me, when I moved to The US this year, I absolutely did not understand the accounting system here at all. So all I did was I got Claude to harvest everything from my Gmail inbox.
11:27I took the floor plan from the house that I'm renting, and I told it to claim any official expense that it possibly can, circle the room that I'm using so it had the exact square meterage that it needed in order to claim the right amount of money from the rent, the bills, whatever it needed. It had all the documents that it needed to do that with the exact calculations that it needed, and what it produced was a perfectly flourished Excel document that I just handed to my accountant.
11:50So it's things like that that make this really valuable with the admin perspective because historically, I used to have to go into every system, pull out the things, and then manually calculate whatever I needed to figure out how much money I could get back from tax relief. So again, in terms of connectivity, we've covered that because admin pulls from the exact same place with our Gmail, our G Drive, and the agents will put them wherever they need to go.
12:09When it comes down to money, things are a little bit different. So again, I own two businesses and I obviously have personal finance. So for me, my dashboard over here has my business stuff primarily.
12:18I don't care about my personal finance because if my business isn't winning, that means ultimately I'm going to fail in life as well. So I don't really track personal finances on this dashboard. I just use an app on my phone, but it's also connected to an app that has a web app, and that's very important for you because we need to piggyback off of their API in order to get our live analytics.
12:35Now depending on where you are in the world, you will have different ways to get your bank information. If you're in The US, there are a few things you can do. I'm using something called Lunch Money, which is this very simple app.
12:44It's like $10 a month. It has full financial information into any of your accounts that you wanna add into here. More importantly,
12:50it has a developer section where we can just pull straight out of their API or use MCP from whatever it is that you've added. The reason we do this is because they take care of the intermediary connection to our bank. Banks don't release things to you, the general public.
13:02You have to go through a platform like Playd. So that is an option, but but they will vet you and they will check what type of company you have and why you're trying to connect an app to this specific thing. So having a middleman can make it a lot easier.
13:13If you didn't want to go down that route, there's something called Simplefin, which is literally just a bridge that I just spoke about. So you wouldn't have all of the analytics in that web app, but it would give you the bridge that you need to your bank accounts. Again, is region dependent, so you would need to check what is available in your area to do the exact same thing.
13:27But really, it's super simple here. All you would need to do is log in, grab your API key like you would for anything else that you're building at the AI at the moment, chuck it in Claude, say this is my API key, I'm using lunch money, I want you to connect and pull down my finances on a regular basis. That would set up polling if you wanted live updates, but you could also just have it on a skill like I do.
13:45If you're coming at this from a business perspective, it's even easier. If you're using QuickBooks or something, it can use the API directly from there and again then QuickBooks would be the intermediary layer between the banks and what you're doing with your own dashboard over here with Claude. In terms of where you would stash it, it obviously lives in your dot EMV file out here and if all of this is completely new to you, don't worry.
14:02I've got tons of videos on my channel walking through the AIOS model as a whole, specifically for business. But this one's for life, and as part of those things, will see the steps that you need to take to build all of this out, including how to understand Claude from a complete beginner to a pro. I will link all that stuff down below.
14:17Furthering on from this, we obviously want to know what subscriptions we have because rocket money used to be a thing and we don't really need that anymore because now we can just see what subscriptions we have based on the direct debits that come out of our accounts every month. This And thing just prompts you, do you still need any of these?
14:30And you would check this regularly to see if you had it. If you wanted to cancel something, you would just click on this cancel button and it would take you to the website to go and cancel your subscription. Here, this thing takes us straight to Adobe to cancel out.
14:40So you can see here again, the goal is take action. We never want to be left with something that is just a static image. All of that's already been done.
14:47We need to take things to the next level which is an operating system. That is the whole point in the word, to operate on what we need to do. And then finally over here, think this is my magnum opus because I have so much social stuff going on that I can't keep up with it.
14:59I have a very needy two set of brothers who constantly nag me for their attention and I just don't have time for that. So what I've done is I've outsourced as much as I can being the best uncle that I am. I never remember anyone's birthdays or what to get them.
15:11So what I've done is I've created separate panels around here. Once we've gathered all the information about the family, it can tell us exactly what they would want for their birthday.
15:19And then at the same time, it can go and research ideas that we can just find and buy. So for instance here, Diane, my imaginary mother, she's turning 64 in six days. She's deep into her garden and never without a crime novel.
15:32An heirloom sea collection paired with the new ton of French around $50. Then on the left, can see what my social life is looking about. I've got dinner with Sam at Bartulia.
15:42All of these amazing places. I can just open up my Google Calendar and it would then show me all of the plans that I have. You can see my week is super stacked.
15:49We can close this for now. But the goal here again is to give us information about what we're doing and what we should bring and maybe even how we should behave at a six year old's birthday party because I don't think taking beer and whiskey would be a good idea to that. So it really fills in the gaps if your social etiquette is not up to par.
16:04And in terms of how this works, again, it's just using a skill like everything else. So for this one, we have got our social skill relationship intel. And down here, it pulls in all the information from the context that we fed it about our family and our extended family and things like that.
16:18And it stores everything into a category. So it would know who these people are. We have all of their information, their children, and that's how it understands to link with our calendar, with our email inbox,
16:28what these people like, and it can then classify our dates, classify what we should be taking to these people. And from there, it can just turn it into whatever it needs to so that you're ready to go for your social event. And then just before we wrap up here, you can obviously add more pods to this if you want.
16:41Like, could have a health tab. I don't track my health metrics. I go to gym every day.
16:44I eat well and my four year old is my blood boy. So you know, I'm doing the basics here. But you could track and add whatever you wanted from your Apple data and things like that.
16:52You just need to have the right apps in place that you can pull down that information from an API like we do. And then you could build an extended dashboard here specifically to track your health if you needed to track certain metrics or your sleep. Just note, like with everything else that you're doing when you're working with AI, it needs to stem from a constraint.
17:07Don't just build something because it's cool. Everything that I've built here is literally things that I use every day and it was built from a massive constraint even more so than the approaches that I take with my business. Then just before we go, I wanna go over a few of the skills that go into each of the pods that we're working with here.
17:21So we got the personal inbox one. That's the one that we looked at in a little bit of detail. It triages the emails themselves, and it forms part of our today notes.
17:29We have the calendar intel that reads the calendar, organizes it, makes sure everything's where it should be. Again, that goes in the today agenda. We then have personal finance.
17:37That is going to sync everything from our bank between the intermediary and our dashboard. It's also gonna scan any bills that we have so that it takes that into consideration when it's planning our finances and our dashboards and things like that. Then we have Document Vault.
17:49That's pretty self explanatory. It goes in the admin tab, and that's pulling out all of our documents from our Gmail inbox. Again, it's very important that you get all of your external systems to send all of the invoices and bills to your central email inbox.
18:01In terms of relationship intel, that is all onboarded with Claude. So you could literally just sit down with this thing and say, I wanna onboard my family and my personal life. This is what it looks like.
18:10And then just tell it about your family, when their birthdays are, what they like, what they don't like. That's kind of what I've done. You can even turn that into a skill.
18:16We then have the action queue and this is for every single tab. This is a skill that just runs through all of these things that it's coordinating to make sure that the right place gets the right information whenever it needs it.
18:25Session prep renders our dashboard for us whenever we have this information stored somewhere, it updates it when we need it. And then finally, we have notifications that are pushed to Telegram via our simple Telegram API integration. There are many different ways that you can do this.
18:37Don't feel like you have to do this exact thing. You can add as many things. You can remove things.
18:41The options are really endless. I just wanted to show you an overview of what this can look like and you can easily walk through this with Claude. Again, I will link some stuff down below that will help you learn how to build skills and walk you through this whole journey from start to finish.
18:53In terms of how to schedule this, you absolutely do not need this to be real time. You could run this at 07:00 every morning. It takes maybe ten minutes at an absolute max for all of these skills to go and do what they need to do, but more likely three to five minutes.
19:05I certainly wouldn't waste one of my Claude routine credits on this type of thing. You can just run it locally if you have a twenty four seven machine, then you can do this multiple times a day if you wanted to. It all depends on how you want to work.
19:15For me, I think something that I will be augmenting is having both the morning brief and then having something that wraps off towards the end of the day that says, hey, did you actually do the stuff that I told you to do today? That way it prods me to making sure that nothing gets carried over to the next day. My whole goal with this system is to hold me accountable to what I wanna do to make sure that I don't have this constant back and forth.
19:34Then we need to talk about the elephant in the room because obviously all of the bigger labs out there are gonna be building stuff like this. We just heard that Google Gemini Spark is coming along. It's released in beta, I think, next week at the time of me recording this.
19:45Currently, they're promising twenty four seven agency system that connects directly to Gmail in your workspace and pretty much does most of the stuff that the admin part of LifeOS does. The problem with that is that it's only for certain users in The US. You have be on a $100 ultra plan, and realistically, you don't need to do that.
20:01You don't need to be locked into someone. You can just build this thing yourself. Secondly, Anthropic is definitely gonna be building the same thing.
20:07We saw in the stuff that they leaked that they are releasing Kairos, which is pretty much the same thing. It's an always on background agent
20:13that's going to be doing proactive check ins, looking for tasks, and then implementing them. Sounds very similar to this, but I'm sure it's going have a few differences. Then finally, we sit on the end here with LifeOS.
20:21This is a custom build. Realistically, this stuff takes like fifteen minutes to build. So if anybody has this great idea that they wanna go and sell this to someone, it's a commodity, guys.
20:30The AI operating system as itself is a commodity. Anyone can build one.
20:34That is not where your value lies in this regardless of what the kiddies out there are saying about the AIOS model as a whole. The value comes in your experience within a specific domain, something that makes you more valuable than the person next to you who can just talk to Claude and get this thing to build anything at once for someone else.
20:50You also have to remember that these bigger labs will outbuild you. They have tons more money than you ever will. So they are already thinking about the ideas that you are going to think about.
20:58So again, don't focus on the tools, focus on the domain expertise and how you can tailor a specific solution to a problem that a client is having. Then one last thing just before we dip off, obviously, Cowork has live artifacts, which is just an HTML face, kind of like what we've built over there with LifeOver. So for the most part, you could build a lot of what I built in HTML, just in the browser, but you could build it inside Cowork.
21:19And then, obviously, your little agent could go and run the skills from within here and present you your live artifact every time it opens up. There are a few nuances with this though because this runs in a sandboxed environment you can't really play with SQLite databases in the same way that I can so as long as your database was in an external system you could store your state there in which case the dashboard would work.
21:37But there are a few little janky things that you need to get around, which is why I didn't set it up for this demo. I think it will make far more sense once Kairos comes out. That's probably where they're going with this whole thing to connect the front end with the back end and have that whole agentic layer.
21:50But who knows? Perhaps in v two of LifeOS, I will release something specifically around Cowork. But that pretty much wraps it up.
21:55You can grab the prompt for this down below. It will build it as best as it can based on your needs. You will obviously need to include specific systems based on your region, but it will definitely give you the scaffolding and everything else that you need.
22:05So I hope this video was helpful. Leave some comments down below, and I'll get back to you as soon as possible. Otherwise, check out the videos on the screen.
22:11They'll definitely help you in your journey. Or you can have a look at my community where we are building the AIOS model every day. I'll see you guys later.
— full transcript
§ 05 · For Joe
Build the system that does the research, not the reminder.
WHAT TO LEARN
An AI agent that only shows you information is just a fancier notification; the one worth building stages a decision and hands you an approve button.
- Design around a real constraint first — a pile of 33 unread Gmail reminders is the right starting point, not a feature wish-list.
- Make Gmail the single intake point for everything: invoices, bills, calendar invites, and task reminders all flow to one place so one MCP connection covers the whole system.
- Separate your skills by pod and give each one a single output destination — this is what keeps a multi-skill system debuggable when one piece breaks.
- Never let the agent act unilaterally on money or communications — staging an action for human approval is a feature, not a limitation, especially in a system you trust with financial data.
- A financial intermediary (Lunch Money, SimpleFin) is the practical path to live bank data; direct bank API access requires vetting most solo builders will not pass.
- The social pod is underrated: a skill that reads your calendar, stores relationship context, and researches birthday gifts outperforms any reminder app for the people you actually care about.
- Schedule the morning run locally at a fixed time rather than burning cloud-agent credits — five minutes on a 24/7 machine covers the same ground for free.
- The commodity warning applies to every AI tool you consider productizing: by the time you have shipped it, a well-funded lab will have shipped a native version. The defensible version is built for a specific domain you know deeply.
