We did not know exactly what cohort 01 would feel like. We had a curriculum, a set of tools, and a strong point of view on how AI systems should be built. What we did not have was the people.
Then the people arrived. And it became something.
Where everyone started
The first session was a welcome call. No code, no demos. Just introductions and an honest question: before AI, what did your work actually feel like?
The answers were good.
One person talked about the joy of Stack Overflow, the dopamine of finding the right answer after real digging, and how that feeling is mostly gone now. Another talked about flow state, hours of writing elegant code, and the identity that came with it. A tester talked about how AI handed her back time she used to lose to documentation and in-sprint automation, how it felt like a superpower she did not expect.
None of these answers were wrong. They were all true at the same time. That tension, between what was lost and what became possible, ran underneath every session we had.
What we actually covered
Week by week, we moved from understanding to building.
The landscape first. How the field shifted from raw LLM calls to RAG to agentic systems. The difference between a model that responds and an agent that pursues a goal until it is done, or until the tokens run out. Why existing skills in system design, version control, and testing do not go away. They just get a new coat.
Then tool calling, orchestration, Crew AI. How you go from a single API call to a crew of agents with roles, goals, and tools. How to route between models based on cost and capability. How to fall back gracefully when a provider goes down, because they do go down.
Guardrails showed up early and stayed throughout. One moment that stuck: someone pointed out that you could ask Chipotle’s customer bot to explain Python, and it would. That is not a model failure. That is an engineering failure. We talked about what it means to actually constrain a system, not just hope for the best.
Then deployment. Docker, Railway, GitHub Actions. Getting things off a laptop and into the world. This is where Keenan, who has been shipping AI products in production for years, was invaluable. The message was simple: if it is not running somewhere, it does not exist.
Observability last, and it hit hardest. The moment the cohort instrumented their pipelines and traced their agent runs properly, the reaction was the same across the board. “I thought I understood what my system was doing.” They did not. Now they do.
The capstones
This is what the whole thing was building toward.
Everyone came in with an idea. Some changed it halfway through. That is fine, we said so on day one. What mattered was that by week six, the idea had become a thing. Running. Observable. Recoverable.
We saw projects that surprised us. The range of what people built, given the same six weeks and the same foundation, was genuinely impressive. Not because the code was perfect. Because it was real.
What we are taking forward
A few things we will carry into cohort 02.
The conversation on identity and loss matters more than we expected. Engineers are not just learning new tools. They are renegotiating their relationship with their craft. That deserves more space.
Observability needs to come earlier. By the time we got there, people were already deep in their builds. Instrumenting a system from day one is a different habit than retrofitting it in week five.
And the community held. People helped each other. They showed up. That was not guaranteed, and it was not nothing.
We set out to teach people how to build AI systems that actually run. What we did not expect was how much we would learn from the people who showed up to build them.
Cohort 02 is already underway. If you want to be part of what comes next, applications are open.