WEBVTT 00:00:04.480 --> 00:00:23.305 Welcome back to How I AI. I'm Claire Vaux, product leader and AI obsessive here on a mission to help you build better with these new tools. Today, I attended Code with Claude, Anthropic's first developer event, and they announced some things in Claude code and Claude managed agents I think you wanna know about. I'm gonna walk you through five things that launched today, 00:00:23.465 --> 00:01:03.340 how they work, what they are, and what I might build with them. We're gonna keep it under ten minutes, and this gonna be a quick preview of what you'll see in your new Claude code and Claude API products experience. K. The first thing that shipped at code with Claude that I think you wanna know about are some updates to the Claude code app. And one of the updates that I know we've all been waiting for is routines, the ability to trigger events or actions on a schedule. We love it. You know I love OpenCLAW, and what I love about OpenCLAW are the crons. And so now ClawdCode has that built in right here in the app. All you have to do is click new routine. You can either run it locally or remote. I'm gonna run it 00:01:03.820 --> 00:01:04.620 locally, 00:01:04.940 --> 00:01:06.620 and I'm gonna say weekly 00:01:06.620 --> 00:01:12.735 newsletter. This is something that I haven't been doing. I'm sorry. If you're a ChatPurity customer, I know I haven't done my newsletter. 00:01:12.815 --> 00:01:17.455 I'm gonna say, look at our change log and draft 00:01:17.455 --> 00:01:18.815 a newsletter 00:01:18.815 --> 00:01:25.535 for us to send weekly. And then I'm gonna go in here and just say, we have a changelog.md 00:01:25.535 --> 00:01:26.415 in the 00:01:27.030 --> 00:01:28.310 docs folder. 00:01:29.110 --> 00:01:31.190 Review it every Monday 00:01:31.190 --> 00:01:34.390 and write a customer facing newsletter 00:01:35.030 --> 00:01:38.870 based on the best customer facing features we shipped. 00:01:40.255 --> 00:01:43.295 Don't talk about behind the scenes 00:01:43.775 --> 00:01:45.935 things like tech debt 00:01:46.095 --> 00:01:46.895 or 00:01:46.895 --> 00:01:47.775 security 00:01:48.095 --> 00:01:51.375 unless they really impress customers. 00:01:51.375 --> 00:01:55.055 Okay. So I'm gonna do that. I'm gonna run it daily or no. Sorry. I'm gonna run it weekly 00:01:55.610 --> 00:01:58.490 on Mondays at 6AM, 00:01:59.290 --> 00:02:00.730 and I think 00:02:01.130 --> 00:02:02.570 that's all I need to do. 00:02:03.610 --> 00:02:05.370 Oh, I'm gonna select my folder, 00:02:05.930 --> 00:02:09.610 um, where my project is, and then I'm gonna create create. And now 00:02:10.035 --> 00:02:13.315 Cloud Code will run my newsletter draft 00:02:13.395 --> 00:02:18.035 cron every week, and then I can come back in here, grab the HTML. 00:02:18.035 --> 00:02:31.450 If I were being really fancy, I would hook this up to my newsletter platform. I would hook it up to my Slack and ping us. But, again, this is very useful to me. This is something that I used to kick off manually in Claude Code every Monday, 00:02:31.690 --> 00:02:35.770 and now I can do that here in Claude Code on a schedule. 00:02:37.125 --> 00:02:43.045 So how does that work? There are three trigger types. You can trigger them on a cron, which is a schedule, 00:02:43.445 --> 00:02:44.645 um, HTTP 00:02:44.645 --> 00:02:50.085 or a GitHub webhook. So you can do sort of a normal webhook or a GitHub webfoot. You can trigger these three ways, 00:02:50.760 --> 00:02:55.000 scheduled, like I just did, off of a GitHub action or a general webhook. 00:02:55.080 --> 00:02:58.440 So you can hook it up to other systems to kick off a routine. 00:02:58.840 --> 00:03:06.120 All the stuff in connectors come along. So I have Slack connected. I have GitHub connected. So you can use those things as part of your routine, 00:03:06.815 --> 00:03:15.215 and it can run-in the cloud or it could run on your laptop like I showed. And this is an example of a use case where you could say weekly, 00:03:15.295 --> 00:03:17.695 I want you to check every PRD modified 00:03:17.695 --> 00:03:24.640 this week and check if it matches our rubric and post a summary to the team channel. So that's item one. 00:03:25.920 --> 00:03:29.600 The second one is in Cloud managed agents in the API. 00:03:29.680 --> 00:03:31.120 If you haven't paid attention, 00:03:31.360 --> 00:03:47.615 OpenAI released something in Codex called goal. You can do slash goal in beta and Codex, and it'll basically bang its head against the problem, do what's called a Ralph loop against the problem until it actually hits the goal. Anthropic released something very similar in the Cloud API called outcomes. 00:03:48.040 --> 00:03:57.000 You define what done looks like for an agent. It can self grade and iterate until it gets there. There's a couple interesting things you need to know about how outcomes are defined. 00:03:57.400 --> 00:04:07.805 They all anchor on what's called a rubric. So there's a markdown file that's uploaded either through the files API or in line, and it's gonna tell your agent what success looks like. 00:04:08.285 --> 00:04:10.685 Then there is a grader, 00:04:10.685 --> 00:04:14.525 and it can do up to 20 iterations on the task to 00:04:15.020 --> 00:04:31.855 get to the outcome that you're going for. I wanna walk through this one in a very specific example to make this just a little bit more concrete for folks. So so imagine that I want you to ship a ship ready PRD. And I don't know if you can relate to this. Often you go through feedback cycles. You have check it against priorities. 00:04:32.095 --> 00:04:34.655 You have to check it against technical capabilities. 00:04:35.135 --> 00:04:39.775 Now using a quad managed agent, you could, in theory, write a rubric, 00:04:40.095 --> 00:04:49.160 which is what does a good ship ready, um, PRD look like. And then the agent can just take your PRD or your idea 00:04:49.320 --> 00:04:53.320 and iterate over and over and over again until it's fixed. 00:04:53.400 --> 00:04:54.440 Of course, you could expect 00:04:55.160 --> 00:05:02.475 of course, you could expose this to your customers in an app like I might do for chat PRD. But, again, this idea of outcome is define a rubric, 00:05:02.955 --> 00:05:13.515 give the agent the task, let it bang its head against that at least 20 times till it gets it right. I think this is really interesting model for agentic products and something I suspect many of us will use. 00:05:14.290 --> 00:05:22.610 The second thing I really love is a multi agent framework supported in Cloud managed agents. So now you can, through the API, 00:05:23.010 --> 00:05:29.090 explicitly define a multi agent team that's going to work against the same container, the same file system 00:05:29.415 --> 00:05:31.735 up to 25, which is kind of amazing. 00:05:32.295 --> 00:05:37.895 You can have a a orchestrator and then delegates, and so there's explicit hierarchy. 00:05:38.295 --> 00:05:46.630 And each agent can have its own toolset. I think this is really cool because now you're able to define not just individual agents, but teams of agents programmatically 00:05:46.630 --> 00:05:48.070 through the API. 00:05:48.390 --> 00:05:53.270 And so the example I would give for something like chat PRD is you could have a PRD orchestrator. 00:05:53.695 --> 00:06:03.295 This is sort of like the master agent that is intended to define and drive the work across the team. And then you can have three pieces or three sub agents, 00:06:03.775 --> 00:06:05.135 a strategy agent 00:06:05.295 --> 00:06:16.790 that reflects the CPO voice, the critic agent that's sort of, like, supposed to poke at the holes in the PRD. I like being the critic agent. And then end review that can maybe have access to something like GitHub 00:06:16.790 --> 00:06:17.510 to 00:06:17.750 --> 00:06:19.830 optimize the technical implementation 00:06:19.830 --> 00:06:20.870 of the PRD. 00:06:21.355 --> 00:06:26.075 And so you can define this as you see over here in the API. 00:06:26.395 --> 00:06:29.515 You define an agent in the API. You give it 00:06:30.235 --> 00:06:34.235 a orchestrator level set of tools, and then you can define 00:06:34.475 --> 00:06:35.595 the sub agents 00:06:36.540 --> 00:06:42.380 in the in the API with their own set of tools. And then you could expose that, as you can see here on the right, 00:06:42.940 --> 00:06:43.580 as 00:06:43.980 --> 00:06:47.740 three agents all working in parallel against the same problem 00:06:47.980 --> 00:06:50.965 owned by the coordinator or level. 00:06:50.965 --> 00:06:52.885 Again, I think this is an interesting 00:06:53.365 --> 00:06:54.245 enhancement 00:06:54.245 --> 00:06:59.045 on the primitives of agents that people are gonna be using quite a bit. 00:07:00.885 --> 00:07:04.565 Okay. The next one I really like, it is dreams. 00:07:04.820 --> 00:07:07.940 So this is all about agent memories. 00:07:07.940 --> 00:07:22.565 Just to make it simple for folks, memory is basically the idea of writing markdown files to the file system your agent uses that helps it do a better job the next time. It's not that fancy. Often those files have a date on them, but you don't really have to overthink it. 00:07:23.045 --> 00:07:32.005 But creating those memories is a little hard. And often a lot of the harnesses right now write memory on a hook. They write them on an event. 00:07:32.325 --> 00:07:40.940 And so what they do is, like, when you close a session, it writes memory or when something happens, write memory. Or, like, with Open Claw, you can explicitly tell it to write memory. 00:07:41.580 --> 00:07:44.780 But what I like about Dreams, 00:07:44.780 --> 00:07:47.420 which is a very funny brand for 00:07:47.675 --> 00:07:50.635 an agentic memory product, but we'll allow it, 00:07:50.955 --> 00:07:51.515 is 00:07:51.755 --> 00:08:00.075 it's a primitive to call against a list of agent sessions. So let's say you've done 50 things with your agent. 00:08:00.315 --> 00:08:01.915 It's an explicit call 00:08:02.170 --> 00:08:04.010 to take those 50 00:08:04.410 --> 00:08:05.370 sessions, 00:08:06.010 --> 00:08:11.370 review them, and then come up with important memories to write to disc. 00:08:11.530 --> 00:08:16.170 And as I'm saying this, I guess this is what we do when we dream. We go through our day. 00:08:16.490 --> 00:08:34.300 We review it silently, and then we decide what to commit to memory. I don't know. I don't know if this is the perfect metaphor, but it's the one we got, and it looks great on a on a branded website. This one's in research preview, so I don't think everybody has access to this through API. I certainly don't have access to it, so I'm looking forward to touching it. 00:08:34.700 --> 00:08:37.660 But why I think this one's important to know is 00:08:37.900 --> 00:08:41.260 it just gives you a frame of reference for how Anthropic 00:08:41.260 --> 00:08:49.865 and these labs are thinking about the primitives, again, of agents and agent memory. And you can predict that some framework like this is going to be integrated 00:08:50.105 --> 00:08:53.625 into agentic platforms or agentic products where on 00:08:53.945 --> 00:08:56.265 some action or some review regular cadence, 00:08:56.940 --> 00:09:15.785 You're gonna review past sessions, and you're gonna explicitly write the right things to disc so they can be referred to moving forward. Side note, I think we think a lot about agent memory, but not a lot about agent forgetting. So I'm looking forward to, like, the purge version of this, which is dreams that 00:09:15.945 --> 00:09:28.870 tell you what to forget. I don't know if that's, like, trauma erasure or whatever, but I think there's something interesting here. Okay. And then number five, the only announcement people really care about, which is usage limits, 00:09:29.430 --> 00:09:34.630 are up. So starting today, Claude Codes' five hour limits are now doubled 00:09:34.870 --> 00:09:38.790 across pro, max, team, and seat based enterprise platforms. 00:09:39.285 --> 00:09:47.285 Peak hours are going away for pro and max plans, and the rate limits for Opus models in the API are going up. 00:09:47.605 --> 00:09:50.085 So we can all use these products more. 00:09:50.405 --> 00:09:55.620 Again, what do we see today at Code with Claude? Lots of other stuff. They might put data centers in space. 00:09:56.020 --> 00:10:02.580 There was a partnership with SpaceX. All sorts of fun things are happening. But the things that I think you can use immediately, 00:10:03.220 --> 00:10:15.455 routines for scheduling tasks in Claude code so you can get things done either on a webhook or on a schedule, which is the ability to set a rubric and task and have an agent work against that task at least 20 times to 00:10:15.775 --> 00:10:16.975 nail the rubric, 00:10:17.375 --> 00:10:23.855 multi agent orchestration, which allows you in the API to define an orchestrator role and sub agents, I think up to 25, 00:10:24.280 --> 00:10:27.880 to get work done from different points of view with different tools, 00:10:28.200 --> 00:10:32.200 dreams, which are a way to consolidate agent memory over sessions 00:10:32.360 --> 00:10:36.920 over time and do that on demand. And then finally, we can all use 00:10:37.795 --> 00:10:42.275 more Cloud Code, which makes everyone happy. I do not know if these limit increases 00:10:42.435 --> 00:10:50.035 apply to Cloud Design. I suspect they do not, but I hope they will. I think they launched a lot of stuff that we can use practically 00:10:49.570 --> 00:11:07.705 right now in Cloud Code and gave me some good ideas for what the future of AgenTic products look like. None of this was mind blowing, but I think it's all really practical. And it just gives you the idea that Anthropic is trying to be the agent platform of choice for builders. I look forward to seeing what you build with it, and I'm gonna go play with the API. 00:11:08.025 --> 00:11:08.665 Bye. 00:11:09.945 --> 00:11:20.740 Thanks so much for watching. If you enjoyed the show, please like and subscribe here on YouTube or even better, leave us a comment with your thoughts. You can also find this podcast on Apple Podcasts, 00:11:20.740 --> 00:11:21.540 Spotify, 00:11:21.700 --> 00:11:23.540 or your favorite podcast app. 00:11:23.700 --> 00:11:28.310 Please consider leaving us a rating and review which will help others find the show. 00:11:28.550 --> 00:11:34.630 You can see all our episodes and learn more about the show at howiaipod.com. 00:11:34.790 --> 00:11:35.590 See you next time.