Anthropic just dropped 5 workshops on building self-improving agentic systems from scratch: 00:00 - Ship your first Claude agent 36:44 - Build memory for Claude agents 1:05:06 - Make your agent autonomous 1:26:46 - Set up a proactive agent 2:03:35 - self-improving agents (tools,skills) These 3-hours of free Claude... workshops will replace 10 paid agentic courses. Watch today, then read article below on how to build a self-improving agentic system with Fable 5.show more

Love it. But here's what I realized: knowing how to use one model really well matters less than knowing which model to use for which problem. That's where the real value is.

Yeah, this is true. I mostly use Sonnet for coding, Opus for orchestrating, Fable for planning and goal setup.

I think a lot of people are doing that. The model choices may be different for each task, but I do think most are splitting their loyalties.

cool now watch everyone build the same todo app agent

It's not just about building agents, it's about agentic self-improving systems bro. That's what this series of workshops is about.

2 hours of clean alpha, its will be nice evening

Yeah bro, one watch will replace 10 paid courses. Defo worth giving it some time.

Memory layers require scalable persistent storage infra.

the memory segment at 36:44 caught my attention, that sounds super useful

Yeah, memory layer is the basis of every agentic system, that's for sure. Btw, have you built memory for your agents before?

Thanks for sharing this!

@0xMovez wow, this is gonna be a game changer for my side projects. especially pumped about the self-improving part. gotta dive into these workshops asap!

thanks for sharing and for timecodes

You're welcome mate. Btw, what timecode caught your attention the most?

just started watching

bro made a self-improving agent and it immediately optimized away the need for him 💀

Ahh!! that's pure alpha

Thx bro ! Happy its catched your feed

Your posts are outstanding!

's new masterclass workshops are a blueprint for self-improving agents, but they highlight a massive hurdle: how do you share these memories across a multi-model ecosystem? If you deploy a fleet where Claude handles reasoning but passes tasks to specialized @GeminiApp or @OpenAI models, trapping user state in model specific, ephemeral wrappers is why production systems break. True cross model autonomy requires a model agnostic infrastructure that treats agent memory not as isolated vendor logs, but as a core, unified, queryable data layer. #AgenticAI #AIMemory #AIInfrastructure #AIAgents

this course is definitely game changer, thank you movez

Yeah, 3 hours of watch and you're at master level on agentic systems.

Step 2, connect the agent to customers

this is exactly what devs need right now

Yeah bro, this is a real roadmap for building self-improving agents. Btw, are you building one yourself?

Love it. How do I do it without Anthropic getting any of my data/ training on it?

W sharing about Anthropic workshops, thanks broski

That’s huge—five workshops is a gold mine. Excited to dive in and see self-improving agents in action. Let’s ship some next-level Claude work.

Five workshops, future looks busy

Free workshops are great, but “replace paid courses” depends entirely on whether they generalize beyond Anthropic’s stack.

Official workshops are often a better place to learn than random AI courses. You get the concepts from the people building the tools

mark and thanks for the sharing

Using AI. You're taking on judgment debt AI removes the small decisions so you can focus on the big ones. Sounds like a feature. It's actually a trap. Judgment is built through reps. Kill the small decisions and you hollow out the intuition you need when the big ones arrive.

Great

This is definitely worth watching.

PURE ALPHA

This video is really worthy of being saved

Anthropic 官方下场出这套教程确实挺有价值，基本把当下 Agent 演进的标准形态给盘清了。 不过上手自己搓一遍就会发现，所谓的 autonomous 或是 self-improving，剥开来看，底层其实也就是扎实的状态管理加上几套 feedback loop。 跟着讲义跑通 demo 挺快，但真丢到真实的业务环境里，真正卡脖子的往往不是怎么让 Agent 自己动起来，而是怎么配一套靠谱的 eval 机制去兜底它的“自由发挥”。现在框架和概念都很丰满，只是落地到最后，拼的其实还是这些脏活累活。

tools plus skills changes everything

That’s a talk from “Code with Claude”, on May 6, before the Fable release.

Summarized some other useful posts about self-improving agents here:

3 hours of this beats most paid courses easily

felt that

The memory and self-improving sessions are crucial. Pro tip: don't just store raw logs. Vectorize past failures so the agent actively avoids repeating mistakes. Learning from errors is the real key to true autonomy.