Happy Operators AI Cohort
Five builders, five stacks, one week: each shipping an agent suited to their own work.
The question I was chasing
Could five people each build the agent their own business needed, rather than something from a template?
Most AI training solves the hard part by removing it: one stack, one worked example, one pipeline everyone copies. That produces five people who can rebuild the demo and none who can build for their own work. Happy Operators wanted the opposite. Five people, at different levels, each building exactly what they needed for their own business, with support and a one-week deadline.
The constraints
The experience levels of the group were the constraint. One member turned up with an agent system already half built and wanted evals, optimisation and someone to hold them to a deadline. Another had never installed a command-line tool and needed a one-to-one before they could run a single command on their own machine. A single programme structure cannot serve both.
It was an intensive week with daily check-ins and async support, so wrong turns and failed experiments could be recovered. Whatever each person started had to be small enough to finish and real enough to matter to their business. That call was made with them on day one.
The decisions that mattered
Personal projects, not a shared exercise. Everyone built for their own business, so nobody could copy anyone and everybody had to make their own calls. It also meant five different stacks in one room. Agents on people's own cloud infrastructure, agents running locally, and agents assembled from off-the-shelf parts like Claude Code inside Claude Desktop wired to Apify.
Two channels, not one. Daily group check-ins for momentum and for seeing each other's problems, one-to-ones for anything that mattered to only one person, office hours and async support in between. Debugging one person's install in a group call wastes four people's afternoon.
A demo day at the end, announced at the start. The deadline did more work than any lesson could have. It set the scope for everyone: not the most impressive thing you can imagine, the most useful thing you can show on Friday.
Built with: Claude Code, Claude Desktop, Apify, self-hosted cloud agents, local runtimes
Where it landed
All five of the group demoed on Friday. Everyone had achieved something significant in their own way. The week kept people accountable, and we had been through an intensive programme together, but ultimately each person's build was their own. What formed was a real sense of community: people working on their own projects at their own pace, making progress inside a small supportive batch.
What I did not plan for was how much of the week came from the group, not from me. Five people building different things in parallel, watching each other get stuck in the check-ins, turned out to be its own teacher. That is the part I would design for more deliberately next time, and it is the one thing a recorded course cannot reproduce.
Part of the Rolling Waves work archive.