Anthropic Just Shipped the AWS of AI Agents (Free Dashboard)

Now Anthropic's been shipping many different features recently from managed agents on April 8th. They have routines and Ultra Plan that were shipped on April 14th. And most people are thinking these are just separate products, but let me tell you this, they're not. This is Anthropic's full automation stack, and nobody's really mapped out the complete picture and the complete road map that I think Anthropic's going down with some of these releases. So there's really five layers that we're going to cover in

Today's video, and today we're going to go deep on two specific ones, which are the cloud layers that recently just dropped, which is the managed agents as well as the routines. We'll take a quick look at, you know, Ultra Plan, and then of course, a lot of you are already familiar with skills, hooks, cloud code, but all these together really form the uh current stack that Anthropic is developing for their AI automation. By the end of this video, you will know

Exactly how routines and managed agents work, when to use which, and why Anthropic is secretly building towards something much bigger than any single feature, and something that a lot of people may not realize. And stick around because not only will I be giving you access to exact templates for managed agents, as well as diving deep into each one of these and how they fit into the whole Anthropic ecosystem, but I also built a fully open-source dashboard that you'll be able to

Get access to called Stride Agents that essentially wraps around this entire stack into one UI. I'll show you throughout this video how each one of these pieces work in the entire stack, exactly how these APIs work in practice, and I'll break down the entire thing so by the end when you actually are using this UI, you'll understand how it's working under the hood. And this repo is fully open-source that you're going to get access to. It has everything from a quick

Start dashboard, chat interface, a taskboard to manage the tasks for your agents, uh the agents where you can see the exact agents that you have within Claude's managed agents infrastructure. We have the vaults, the skills, where you can go over here to the skill library and access literally thousands of skills right here where you can easily just one-click install them. You can of course import from GitHub any skill that you want as well. Without ado, let's dive right in. All right,

So before we dive deep, let me just show you a quick overview of the full picture. So the five layers from local to cloud. We have layer one, which is Claude Code. So this is interactive, you know, you sit in the terminal and pair program. Uh then we have layer two, which is skills and hooks. This is still local. You can have this in the cloud, of course, too. Say, for example, in the UI, which I'm going to show you. Like

I said, you can have access to pretty much any skill you want. We have routines, so this is the new uh Anthropic release, which is cloud automation that runs while your laptop is closed. Therefore, we have Ultra Plan, so this is cloud-based collaborative uh planning. And then layer five, which is super super huge, uh which, you know, is still very early in development, and I I am betting that we're going to see a lot of new releases for this, which is

The managed agent. So this is full production agent uh applications on Anthropic's infrastructure. So we've been seeing over the last few weeks to months even where Anthropic is essentially just been cloning uh open claw, but they've been making it in many ways a lot better, especially for their enterprise-level clients who very much value security and whatnot. So in today's video, we're mainly going to focus on routines and managed agents because those are some of the two new ones that, you know,

A lot of people aren't utilizing to their maximum capabilities. Plus, we'll take a look at Ultra Plan as well as some of these other ones briefly. Let's first start off with routines. Now, one thing I will say is this entire GitHub repo to the full Stride Agents open-source repo that we're going to dive into, which handles all of this and under the hood, as well as this entire 20-page document, which goes over everything that I'm showing in this video. If you

Want to dive deeper, you'll be able to get access to all this for free by simply join our Stride AI Academy. So you can join the link down below. All the free resources in today's video will be available there. Routines launched on April 14th, and this feature is essentially what turns Claude Code from a tool that you talk to and into an employee that actually works while you sleep. These are essentially like cron jobs that you can schedule. You can run

One time. You know, this is really just the start, and I'm definitely know we're going to be seeing a lot of development with routines coming soon. A routine is a automation where you configure once, you define the prompt, the repo, the connectors, and it runs on Anthropic's cloud without you having to watch, schedule, or anything like that. So each routine runs as a full Claude session in the cloud. You can have different things like commands, skills, connectors. You know, it's all

Autonomous, and if you've used something like {slash}schedule in the Claude Code CLI before, uh this is kind of similar to that, and this is scheduled routines, but now in the cloud. So it's the same concept, but it's potentially going to be a lot bigger of a feature set once it even gets a little bit more developed. There's really three ways that you can actually go ahead and trigger a routine. The first trigger is scheduled. So this is like a set cron

Job or expression, and the routine basically fires on a schedule. So every night at 2:00 a.m., it pulls the top bug from your issue tracker and attempts a fix or opens a draft PR or does some sort of research. Or every Monday morning, it scans your merged PRs from last week or flags any stale documentation, updates PRs, whatever the case may be. So this essentially replaces the whole leave Claude Code running on your computer overnight, you know, what we were kind

Of doing to try to hack this all together. Now it can all be done in Anthropic's cloud. The second trigger is through API. So each routine gets its own HTTP endpoint and off token. So as you can see, we have that schedule trigger right here. If I go to schedule, I can see that I can schedule it for hourly, daily, weekdays, weekly, custom, etc., right? That's the schedule trigger. The second trigger is the API, right? So I actually have this one

Connected through API. And you'll see each trigger when you do API gets its own URL right here that you can call as well as its own off token. And then you can go here to see an example curl what that would look like. This is essentially just one post request, and then you get back a session URL. All right, so with this API, you can get very creative in terms of what types of routines you can set up. And then the

Third trigger is GitHub events. You can subscribe a routine to a repository, and then when a PR is opened, you could run, you know, a bespoke code review with your team-specific checklist and leave inline comments on any release that you may publish to that repository. You could generate a change log and update the documentation. So as you can see, you can get pretty creative with these triggers and how they can go into the routines that you set up. Now there are

Some limits with the daily runs you get for the routines. So on, depending on what plan you're on, you could either get five runs per day on pro, 15 on max, 25 on team or enterprise. So I'm going to show you how you can actually go ahead and create a routine here. So in your Claude, you'll have to go to claude.ai/code/routines. So if you're in Claude Code, you'll see the routines button here. You can click on that, go to new routine

Up here. So where you can enter the name for your routine. We'll just do YouTube test. And then I can just say do AI research on what's trending today. This is very basic. Like I said, we could get pretty in-depth with setting these up. You know, you can select a repository here. We can add specific connectors if we want. So I could, you know, add Fireflies or Gmail or literally any connector that you already have or add more connectors right here.

I'm just going to remove these for now. You can change some permissions here if you want, and then we can of course go ahead and select the trigger. So for this one, I'm just going to do API um cuz we will be calling this. And I can click on add trigger. Now when I go ahead and click on create, it's going to give me this token right here, which I'm going to delete after this. This is where we can call it

Programmatically through any sort of API, through any specific app, which we're going to do right in our Stride Agents app right here, which like I said, you'll be able to download this and get it set up on your system, which I'll show you how to do as well shortly. But if I go over here to add routines, now the one limitation I would say right now with routines is pretty much have to create it manually through the actual UI, and then

You can call it programmatically um via API. We pasted in that bearer token right here, and I'm going to go ahead and grab the routine idea ID, which is right here in the URL. And then I'm going to come back here, paste in the routine ID. We'll just call this YouTube test. And really, all I'm going to do is I'm just going to change this to trigger type for API. Now you can, like I showed you, uh if you wanted to,

You could set up your trigger to be scheduled like this and have this, you know, daily or hourly at a specific time within the Claude UI. You could also set up a schedule within the app here. Now if you are using it within the app, you there is a few limitations. So you need to be running this npm or pnpm run dev uh and then colon scheduler. Or you could actually run pnpm dev colon all. All right, this will essentially start

Next.js. This app, this open-source tool is built on Next.js, and then it's also using a cron scheduler right here. Now you would have to have this running on your computer for the uh crons to actually work. And then two, if you're running the all command or the tunnel command, it's going to start a Cloudflare tunnel as well. And we'll get into this in a second, but this is actually how we're able to connect a routine to our actual agent for it

To be able to call that API endpoint for the Claude Cloudflare tunnel in order for the routine to be triggered from any agent. I'll just set this to run every hour on the 50-minute mark right here. If you don't want to schedule one, you could just simply toggle this off. And you can change this afterwards, of course, too. But now we're just going to click on add routine. Now we can see our routine right here. You could fire a test right

Here for it to run. And you can also see here that the next fire is in 2 minutes. You can go to calendar here and we can also see any specific routine that is scheduled. So, now that we have this schedule, it's set to run at 4:50, 5:50, 6:50, etc. Now, when would you want to use the scheduler in the Stride Agents app opposed to the actual Claude app right here? Maybe if you just want tracking in the app, but if

You do want it to run on the cloud when you're not running Stride Agents, then you would probably want to set it up actually within the app here. This means in about 1 minute from now we should see this fire, so we'll come back and check that when it does. All right, so it's 4:50 now, so we can see that we have our YouTube test fired right here. We can see the session right here. So, if I clicked on this, it

Would actually go to the specific session in Claude right here for Claude code. And we can see that it is doing exactly that. So, it's doing its research right here. You can see the web search tool. Of course, this is a very basic routine. You could create much more advanced routines with different connectors as well as a much better prompt. And we can see that it is running right here. If we go back to some of our other routines, this is

What you'll see when you have multiple runs for a routine. We can see the different runs, the time as well as when it was actually triggered. So, was it through API right here or was it through manual or webhook or whatever the case may be. And here we can see our output for AI news. Like I mentioned, depending on your subscription level, you do get a certain amount of routines a day and they are based on your Claude plan. So, it's

Going through your subscription. It's not going to be pulling from separate API use. Now, if you do reach your limit, then it would actually pull from extra use. So, branch safety is built in. So, by default routines can only push to branches with Claude {slash} prefix. Now, you can override this per repo, but the default prevents accidental pushes to main. And everything the routine does appears as you. So, commits, PRs, Slack messages, it's your identity, your responsibility. So, at the moment

Routines do belong to just your own individual account, so they're not shared with teammates or anything like that. Like I showed you, there are currently some limitations with routines, you know, with the daily limit as well as, you know, not being able to share it with specific teammates. But this is still a genuine replacement for many different cron jobs you may be running. And the way Anthropic is shipping and building right now, I think, you know, we're going to see a

Lot of these limitations disappear pretty soon. Like I said, this is still in research preview. Routines handle a lot of the executions, those cron jobs, etc. What about the planning before the project actually starts? Well, Ultra Plan can come in handy for that. So, Ultra Plan basically shipped the same day as routines and it solves the problem of collaborative planning before execution and it's all done on the cloud. You can read more about Ultra Plan in this document right here. I'm

Really going to be focusing on routines as well as managed agents cuz those are really where everything ties into the Stride Agents system and what we're really focusing on here. So, I'm just going to skim through this, but you can check that out. It is a very cool feature as well. And I think all these features will continuously be developed upon and start to inter rate a lot more as, you know, weeks, months go on, etc. Managed Agents launched April 8th,

So 6 days before routines came out and I need to be clear about these ones. Routines are essentially to automate your developer workflow, so they run as you, they push to your repos, fire on schedule. That goes through your subscription billing, whereas managed agents are production infrastructure for building agent powered applications that basically serve your users and all of this is going through their API billing, so pay per use. The way I would think about this is if you're automating your

Own personal code review, you can use routines. Now, if you're building a code review product for customers, you would want to use something like managed agents. All right, so now that we know what managed agents are, I'm going to show you how you can set them up. There's a few different ways you can set them up. So, over here we're in the Claude console platform. Now, you'll see they pretty much have this quick start guide where you can chat with Claude

And build out your agent. They also have some templates right here. So, blank agent, deep researcher, structured extractor, field monitor, whatever the case is. Let's say I wanted a support agent, I could simply just click on the support agent template and then we'll see a YAML file or a JSON file. You could use either format here for building your agent. So, let's say I wanted to use YAML, let's go ahead and use this template. And we can see the template's created,

Then we can go ahead and configure our environment. We can decide if we want to use reuse an existing environment or if we want to create a new one. So, let's say I just wanted to reuse one of these. Now, you can see it's going to go ahead and start a session and once it starts a session, it's going to ask well, it's asking us now. We have to connect our specific credentials for Slack and Notion. It's asking us what channel

It should post to, so we can say the support channel. Asking us where our knowledge base is in Notion. I can just say the whole workspace. Now, it's asking us to connect those specific MCP servers for the MCP server for Notion as well as the MCP server for Slack. Now, it's asking which vault it should use for to store these credentials. So, credentials are stored in vaults. You can either select one we already have or create a new one. I'll just

Select this one right here, so our support agent vault. And then this is where you would authorize this MCP server. And we can select our Notion right here and continue. And now we can see that Notion credential is created and stored in our vault. And we just wanted the same thing for Slack here. And allow. And then boom, both of these credentials were created and stored in our vault. All right, so now we can go ahead and do a test run

Here and we will see that our agent has been saved and it's awaiting our first message. So, I'm going to go ahead and say something very simple, what company is this about? You can see that that was our user message. We can see the transcripts here as well as the debug. So, we can see all our request has started, it's thinking. We can see literally all the different logs right here. Also, too, if we go over to our sessions tab here,

We'll see this session has been started. We can see all the different traces. And you'll see here the agent responded saying, "Based on what I can see, this is a port setup for Executive Stride, a company by Josh Pocock, founder and CEO, etc., etc., etc." The cool thing is if we go over to our actual app here and we go over to sessions, we'll see that exact same session was created right here 1 minute ago. If I click into this, we

Can see the exact same info, right? We can see the transcript over here, the debug with all the different run info, what the agent replied right here. Now, we could also continue this in chat, which we'll get into later. You can see all the statistics up here. You can replay this session kind of just to see the speed of how everything went right here. And if we go over to our agents tab up here, this is where we can see this

Most recent support agent that was created. If I click into it, this is where we can see the agent details, right? So, we can see the MCPs and tools with the built-in tools right here. So, bash, read, write, etc. We can see the Notion and Slack credentials for the MCP that are here. You can see the skills if they were attached as well as you can see some additional things that we'll get into in just a second, like the default environment

As well as the ability to attach a vault like so as well as the ability to run the agent on a schedule with a specific environment as well as trigger it through API programmatically. And you can even attach routines to it as well. Now, if we go to the edit section here, this is where you can edit the agent in the form view, the YAML view, or the JSON view. So, typically I would use the form view. You can see the

Different models here, the speeds, so standard, high, system prompt, all the different tools. You can add the MCP server and change this to always allow tools or request confirmation. You can also add metadata here as well. Now, if I click browse skills right here and taking us to our skills area. This skills library alone is super valuable. This is where you can see all your different skills here, the installed skills, bundled skills, Anthropic skills. And you get hub skills, which you

Can import right here through any GitHub URL, as well as a library of thousands of skills, which you can just go, scroll here, search through and install these in one click. So, we got the SEO audit, we got agent browser, front-end design scale. Literally any skill you can think of is pretty much here. If I wanted skill creator and agent browser, I could install these two skills. And now we see that they're installed. I can go over here to my installed

Skills and what do we see? Agent browser right here, which will load up the entire skill with the, you know, read me, all the different folders and files that are in here. We can see the agent skills right here, front-end design skill right here, skill creator right here, so pretty robust. And then I can come back here to our agent and I can simply add as many skills as I want. I got to add agent skills, agent browser, front-end design skill,

Skill creator, and then simply just go down here, save as new version. Now, we can see we have four skills attached. If I wanted to remove some, of course, I could do so. I'm going to show you a few more things here. If we go to the form here, we can see we also have over here MCP servers where you could add from credentials that are already in your vault. So, let's say I wanted to add the ClickUp MCP server right

Here. I could simply just click this and it would add it in with that specific MCP info. Now, if you add a routine to your agent, it would automatically add the Stride Routines MCP right here, which would allow the agent to basically access any routine that you add to it. Like I said before, routines do have some things and limitations that you may want to that wouldn't be of use if you're using this in like a production application or if anything

With an end user. You know, routines are really great right now for personal use, but if you're limited like to 15 a day or 25 a day, then that could be an issue. So, we do, like I showed you, have a way to connect routines to your agent. You simply just add a routine like this, YouTube test, and then we can connect it. You see, once I added that routine, if I go back up here, we can see that our tunnel

Has been connected for this MCP. This is that Cloudflare tunnel that automatically starts if you're using either the TNPM uh tunnel command or the PM PM full command that I showed you. And it automatically added this routines right here, which is pointing towards this Cloudflare tunnel. That's how you can give your uh agent access to a routine. If you add a routine to your agent, it's also going to add this to the system prompt, as well as this, telling it what

Available tools it has for the routine. But at the same time, like I said, there are those limitations. So, what you can actually do as well, which we set up in here, is you could There's a couple options. So, uh when you're using your agents, I'm going to show you in just a second, you can use them in chat or task, and you could override any specific default environment or vaults that you may have connected to the agent. But you do

Have the option here if you want to connect a default environment or vault to your specific agent. Now, you can't actually do do this in Anthropic's. Um it's just connected at runtime. Um so, if I go to a specific agent, you'll see there is no way to add like a default environment or credential vault. But what we can do, when let's say we add this vault right here, we can go over here to schedule an API trigger. So, if I wanted

To run this agent on a schedule, similar to like a routine, let's say I wanted to run this on the hour at 20 minutes, cuz it's 7:18 right now, I could choose an environment here. This would actually override whatever default uh environment is here, and you could select an initial prompt. I could just say do AI research for today. I could either decide to run this now or create it as a schedule. Now, you can, just like routines, also trigger the

Agent through an API uh programmatically if you wanted, and you have the option here for a curl post request right here, and you could do pretty much just follow this. But let's say I wanted to create a schedule. Now, we have a schedule. I can save schedule. We see the schedule is saved. The schedule worker will pick it up within 30 seconds. If everything goes well, in about 1 minute, we should see this agent actually run. So, this is kind of

Like routines, basically, where you can use cron jobs, whatever the case may be. The only difference is, like I said, with routines, you're using your Claude subscription usage, but you're limited. With managed agents, you're using your API tokens, so it's usage-based, but you have less limitations. If we go ahead look at our logs right here, we see that our agent was just triggered right here. So, if I click this, it will open that specific session, and we can see that right

Now, it is running. We want to take a look at our sessions over here in our app. Okay, so I actually encountered an error here, basically cuz the tool had always uh ask permissions on, so it's waiting approval. So, actually, we're just going to go back over here to our agent settings. If we go to edit here, you're just going to want to make sure that for you review confirmation, we're going to want to change it to always allow tools for

These two tools right here. Now, if we go ahead and save right here, now we have our agent updated. I'm going to change this to run every hour on the second minute right here, since it's 8:01, and we should see this run schedule shortly. If we take a look at our logs here, we'll see that this was just triggered for our agent. If I open this session, we'll see the session is currently running. You see that it actually is going, doing

The proper tool calls, and it's allowing all the permissions, of course. And you can see that it's giving us our output right here. So, I cannot find a defined AI research workflow in your Notion. Blah blah blah blah blah, right? Obviously, cuz our Notion doesn't have anything specifically for that, but you would give it the proper tools. You want to actually to do research, maybe give it access to Perplexity, MCP, or parallel, or whatever the case is, or you could simply

Just allow Claude to use its simple fetch tool to do so. So, what we could do over here with our session is we can click on continue in chat if we wanted to, and this will actually allow us to continue in chat. You could also click on actions and then send another chat within that session. But if we're continuing in in chat here, you'll see our entire chat. We'll see what our initial query was. Um we can see really just everything.

We can see the tool calls that it did with Notion, with Slack. So, we can see what the actual input was for those tools, for the query, what the result was from those tool calls. Then we can see the output here, and then we can of course continue on the conversation if we want. This is the chat feature. You could also go to new chat and trigger an agent anywhere, anytime from here. So, you could choose a specific agent. Uh and

Then you could also choose an agent environment right here. Now, if we take a look at our agent environments right here, is where you can create an environment as well. So, you can create an environment in the Claude platform, of course, but you can also do the same thing right here. Now, when you go to create a environment, we do have some templates that you could easily just click to kind of load in. So, now if you want to add a

Specific credential, say like in the actual Claude platform, you can go over to credential vaults, click on add credential, and then you could go down, scroll through, see if there's any one specifically that you want, or you could scroll all the way to the bottom and add a custom MCP server. Let's say for a research agent, I want to use Tavily MCP. I can go ahead and connect it right here. Boom, now we have our Tavily MCP. I go back to

Agents, I want to create a new agent. I could either create an agent actually right here, or what I could do is I could go over to here. We also have a quick-start method to easily create agents as well. We have some templates here. Like so, let's say we wanted the deep researcher, we can go ahead create the agent right here. Now, we could either basically select a limited or unrestricted environment, or create uh one of the ones we already have.

I'm going to go and add and click on our researcher agent one. All right, so then we can actually click on start session and test it over here in the wizard, or we could go over to the chat or task board to test out the agent. So, I'm just going to go to the agent here. I'm going to go to this deep researcher agent. I'm going to click on edit, then I'm going to scroll down to MCP servers, and make sure

That we have our Tavily MCP server, which is right here. And then I'm going to just going to click on save new version. Now, if I was actually doing this, I'd probably add some sort of deep researching skills as well to give our agent a bit of an edge. Now, of course, we could add different things, routines, or, you know, schedules, or really anything like default environments or default vaults, but I'm just going to go ahead and test this in the

Task board. We can go ahead here, and this is essentially just a Kanban board to visualize all the tasks that you create for a specific agent. I'm just going to say do some AI research for today to see what's trending. We're going to use our deep researcher agent, as well as that research agent environment, then create that task. You can go ahead and click on run here, and you'll see that it's going to start to run. And we can see that

The task is running, session is running, and we can see that our uh you know, price right here for our cost is going up. We can see the kind of logs that are being that the agent is using right now. So, using the Tavily search tool multiple different times. If you go over to our sessions either here or in Claude, we'd actually be able to see what it is doing right now. I can see this session right here is running, and

We can see that it has used multiple different uh tools right here with Tavily for searching up AI news, etc. You can see the same exact thing in our app over here. And in the Kanban board, we can see all the different results from the tools, the entire session log going on right here as it's in progress. You can actually just click view session right here to open it up in our tool. Okay, and boom, now we have our session done

Right here. So, if I go click on view session, we can see the session. If I click on continue in chat, which I could also do from the Kanban section, continue in chat, this is where we see the entire research that it did. We can see the tool calls right here. And then we can see the output right here. We can see the frontier model wars at peak intensity. We can see GPT 5.4, Opus 4.7, etc. Pretty much everything regarding AI

In a nice report with all the different references here. If we set it up right, you're going to get a lot better output, of course, with the proper tools, the proper agent, the proper environment, all that good stuff. Of course, you could add skills, as well. And think about that. You could set up multiple different agents. You could have the cron job schedules set up for different time periods, etc. You know, you could even host this on an actual VPS or

On an actual server somewhere, so that, you know, you don't have to have this just running on your local host. The cool thing about Anthropic's architecture is it's essentially new, and it's replacing a lot of the old AI agent frameworks. So, essentially, a lot of the old frameworks treated the agent like a pet. So, one container, session, one harness, and a sandbox all fused together. And if the container crashed, everything was lost, the session, the context, the in-flight tool calls, all

Gone. Anthropic really flipped the script with this, and they treated the compute like cattle. So, it's stateless, disposable, and replaceable. And they isolated the one piece that has to survive. They split every managed agent into three independent components. So, this architecture insight makes everything else really just come together and work. First off, we have the brain here, which is essentially Claude plus the orchestration harness. And this decides what to do, which tools to call, how to interpret those results. So, if

A brain instant crashes, a new one spins up and continues from the session log. So, no data will be lost. This is why horizontal scaling works. Next, we have the hands. So, this is sandboxes, tools, environments, bash, web browsers, MCP servers, custom tools. You know, each hand exposes one interface. So, execute name, input, and return a string. And then, when a hand fails, the brain treats it as a tool error, provisions a new one, and keeps going. The credentials never touch

The sandbox. And then, the session. So, this is an append-only event log. Every tool call, every response, every decision. Sessions persist outside of the brain and the hands. So, if everything crashes, the session log survives. A new brain will call the wake up session ID and replays the log to reconnect full context. Now, this is where the security payoff comes into play, and this is where we don't really see in some tools, you know, like open claw and whatnot. Maybe why

A lot of enterprises may not adapt some of them for security reasons. But, with these managed agent systems, uh the credentials live in the vault at the brain layer, but they never cross into the hands. So, a compromised sandbox cannot leak your API keys because your API keys were never actually there. Pretty much everything I've shown you here is either live or in research preview. But, Anthropic's road map definitely tells a bigger story, and I hope me piecing some of these

Together and showing you some examples of how to actually use this within an app can maybe show you where we're potentially going. So, multi-agent coordination for one. So, for example, a parent agent breaks work into pieces and delegates it to specific child agents running in parallel. Outcomes. So, this would be able to define what success looks like. So, did the PR pass CI, or did the report contain the required sections? So, this would be automated evaluation. And then, memory as well.

So, agents being able to retain knowledge across sessions. So, the agent that reviewed your code base last week remembers the same architecture for this week. Right now, routines are still in beta. You know, they really, like I showed you, only support three trigger types. But, the trajectory is clear with more event sources, more triggers, more automation surface. So, if we look at the patterns we've been seeing with 2024, you know, you own the model. 2025, you own the developer tool, cloud

Code. And then, now 2026, you know, you own the runtime, own the agentic harness. So, each year we're pretty much moving up the stack, and it's probably only going to get, you know, quicker and quicker, maybe each 6 months, or whatever the case is. So, this is really just the AWS playbook that we start saw. You start with compute, add storage, add networking, security, monitoring, and before you know it, leaving essentially means rebuilding everything. So, Anthropic is running the same play

For AI agents here. I personally believe the teams building or at least leveraging this stack today uh in some capacity will have a head start when multi-agent outcomes and memory become, you know, available. The question really isn't whether to use this stack or start implementing in some capacity. It really is which layer do you start with? Like I said, the Stride Agents starter kit that I built for this will all be available in our free school community. So, this should really

Just give you a head start. I did spend a few days really refining this and tweaking it. And if you guys want uh me to continuously still update it, and you do find value in it, or you want to contribute to it yourself, you can definitely do so, and we can definitely build something around that if you guys want. But, yeah, that whole GitHub repo, as well as this entire 24-page document diving a bit deeper into pretty much everything we talked

About here today, will be available for free, 100% in our school community, as well as all the other free videos and resources that we've done on this channel. So, if you're not already in the Stride community, Stride AI community, make sure to join there. You can connect and network with other like-minded AI entrepreneurs, developers, whatever the case is, as well as myself, too. So, if you have any questions, you can ask in there. Other than that, guys, that's pretty much it

For this video. I really hope you got some value here. Let me know in the comments down below if you've been using the managed agents, the routines. But, other than that, guys, I will see you in the next video. Keep hustling. Keep grinding. And of course, guys, accelerate your stride. Take care.