OpenAI just launched a 3-hour course on building agentic systems from scratch: 00:00 - Building workspace agents 37:42 - Deep dive into OpenAI AgentKit 1:24:01 - Agent memory (loops, context, evals) 2:21:44 - Agentic training (OpenAI playbook) This 3-hour watch will replace you 10 paid agentic courses. Watch it... today, then learn how to build a team of AI agents from scratch in the article below.show more

Multi-agent systems don’t fail because they lack agents. They fail because they lack shared, trustworthy context. Without persistent memory and provenance, a “team” of agents is just several black boxes repeating the same work. The real production stack needs: shared context verifiable sources coordinated memory auditable decisions That is why @origin_trail DKG V10 and ethereum:0xaa7a9ca87d3694b5755f213b5d04094b8d0f0a6f matter. More agents are not the moat. Trusted coordination is. @0xCodez @origin_trail @TraceLabsHQ @woisau1 #TRAC #OriginTrail #AgenticAI #MultiAgentSystems #AIMemory #VerifiableAI #KnowledgeGraph #AIInfrastructure

was gonna watch a movie today but looks like spending time on this course is way more productive thanks for sharing

It’s definitely not a movies. Top tier course on agents in my opinion

Oh, a new alpha video from Codez, thanks bro, i will watch it today

Yeah defo worth give it 3-hours. Legendary watch

Never forget to include one agent as a senior architect who has seen it all and plays a devil's advocate with the findings of his colleagues, who works with the KISS principle and thinks that the best work is the one you don't do (YAGNI principle). (YAGNI can be strict)

the agent memory section (loops, context, evals) is the part most people skip and it's the whole ballgame. anyone can wire one agent call. keeping it coherent over a long task is the actual hard problem. thanks for the timestamps

book,book and book this gem

Wow, I haven't seen this video before-it looks like it's worth watching.

today my evening will be busy cause i will learning this course from Anthropic

3 hours is generous, most agent tutorials show 20 minutes of demo and zero minutes of what breaks in production

hey codez, check your DMs - paid campaign

bookmarked, thanks for this alpha

The agent memory section is the one I'd actually pay for. That part usually gets glossed over in these breakdowns.

The memory & evals section is key. Reliable context needs rigorous retrieval evals. For agent memory, pairing deterministic context scoring with strict deduplication prevents duplicate hits from inflating your memory scores.

Strong resource. The part most teams underestimate is the operating layer around the agents: memory, tool boundaries, evals, handoffs, and clear human approval points. That's what turns "agentic" from demo to workflow.

a 3 hour course gets you the vocabulary. what compounds is shipping one agent into your own codebase and reading its failures every day. evals, memory, loops only click once they break on your problem, not on a course repo.

That AgentKit course is useful because it treats memory as part of the loop, not a feature checkbox. The operational question is where that memory should live when work spans more than one coding agent. I wrote a short piece on the local, inspectable version of that boundary:

damn that's a gold vid, thanks for sahring

bookmarked, first agent to production crew in one watch, wild

The evals section at 1:24 is where most people should start, not end. Building the agent is the easy part — knowing whether it actually worked on 500 edge cases is what separates a demo from a deployed product.

Courses teach you to build agents. Nobody teaches you which business problems are actually worth automating first. That gap is where most teams waste months.

covering loops, context, and evals together instead of separately is what actually maps to how production agent systems get built and debugged

very good alpha, thanks

先把收藏夹再塞满一点

that agent memory section at 1:24:01 probably solves tons of real world persistence issues