Give every team one AI hypothesis instead of generic training
Generic AI training across 300 people produces compliance, not adoption. One written hypothesis per team, tested on a retro cadence, makes the rollout correct itself.
Written by
-
Benedikt Stemmildt Founder & Co-CEO
Published: 18 September 2026 Updated: 1 min read
An enterprise group I worked with this month was about to roll AI out to 300 people across 30 teams. The default plan was the obvious one: book the training, run everyone through the same two days, move on.
Generic training produces compliance
We did not start there. Generic training at that size produces compliance. Everyone attends, everyone nods, and a quarter later the daily work has not moved. The training answered a question the people doing the work were not asking.
One hypothesis per team
So each of the 30 teams wrote one hypothesis instead. Not “does AI work” as a yes-or-no for the company, but a specific claim about their own context: here is the thing we believe we can do better with AI, here is how we will measure it, here is when we will know. The team owns the question, so the team owns the answer.
A hypothesis needs teeth
The anti-pattern is the safe one: the team that picks “we have a documentation problem” because everyone already knows agents handle docs. Nothing is learned. A real hypothesis is one the team is not sure about. It can come back false. That is the point.
What holds the program together
- A physical handout at kickoff, so the hypothesis is written down and visible, not a slide that scrolled past.
- A retrospective between every wave, where teams adjust based on what the last weeks actually showed.
- A closing session where every team presents its result to the others, so 30 separate experiments become one shared body of evidence.
Mandatory, not a volunteer pilot
One more piece, and it is not optional. The program is mandatory, framed as the whole organization moving together, not a volunteer pilot a few enthusiasts opt into. Mandatory plus a real retro cadence is what makes the adoption self-correcting at scale.
If you handed your teams one AI question each, how many would pick the safe one?
Further reading
Written by
-
Benedikt Stemmildt · Founder & Co-CEO
More on Adoption
- Engineers finally get to work on the system When an agent gets it wrong, the fix is a rule, a skill or a hook rather than an edit to the code, so the engineer's job moves from working inside the system to working on it. Engineers were rarely given that job and have few criteria for doing it well, so a team has to decide what the work is judged by.
- Find where the work piles up before you pick an AI playbook Four engineering organizations put AI on four different problems. What transfers from them is the diagnostic they all ran first: where does the work actually pile up?
- Run the migration you keep postponing as the training Train agentic engineering on the legacy migration the team has postponed for years, not on a sandbox. The team learns on its real codebase while the debt shrinks.
Scaling needs named owners
Coding agents scale across teams when named people own the platform, the team practice, the guardrails, and the measurement. Rollout & scale is where that ownership is built and handed to internal owners, in your stack.
Training on this topic: Agentic Engineering for Engineering and Product LeadersWant to check one team first? The self-check: twelve yes or no questions, three minutes, instant result, no account. Take the self-check