I ran an AI onboarding workshop with 2 POs and 2 developers. By the end of the week, both POs had opened their first pull requests.
That is the whole story.
Not “people were excited.” Not “the workshop feedback was great.” Not “we had interesting discussions about prompts.”
Two developers came out saying: with this method we can move a lot faster.
And both product owners crossed a line that usually stays untouched for months.
They did not just watch. They did not just ideate. They did not just write tickets a bit more efficiently.
They shipped code into the real workflow.
Lucia Morrone said it better than I could:
“Ich platze vor Stolz. Das Ergebnis ist grandios.”
That is the metric.
Excitement Is Cheap. Crossing Role Boundaries Is Not.
Most AI onboarding still optimizes for the wrong signal.
People leave with ten prompts, a few screenshots, and the vague feeling that something important happened. Then Monday arrives, the real system reappears, and the old role boundaries snap back into place.
The PO writes requirements. The developer implements. The translation layer stays intact.
If that is the outcome, the onboarding did not fail because the model was weak. It failed because the operating model never changed.
The interesting moment is when a person who normally lives upstream of implementation suddenly produces a first artifact that enters the engineering flow without theater around it.
That is what happened here.
We Did Not Install A Prompt. We Installed A Runtime.
The workshop did not revolve around “best prompts.”
We installed the Keeper Runtime for everyone. If you want, call it Keeper OS. The point is the same: this is not a clever text snippet. It is a working environment for the relationship with AI.
Each person set up an agent with:
- its own identity
- its own soul
- its own way of working
- explicit quality expectations
That part matters more than most people think.
Generic AI setups always fail at the same place: they ask five different people with five different working styles to adapt to one vague assistant personality. That creates friction immediately. One person wants challenge. Another wants structure. A third wants speed with hard guardrails. A fourth needs the system to slow down and explain.
If the relationship feels off, people do not build trust. If they do not build trust, they do not use it on real work.
So we flipped that part.
The agent adapts to the person. Not the other way around.
If you want the underlying model, it is on my site as Keeper Runtime.
Personalization Was Necessary. The New Template Made It Practical.
There was one improvement this time that mattered a lot.
I now provide a template folder for the setup.
That sounds small. It is not.
The blank page is one of the biggest hidden taxes in AI onboarding. Even motivated people lose momentum when the first hour turns into “where do I put this file, how do I structure this, what belongs into identity, what belongs into work mode, what is too much, what is too little?”
The template folder removes that tax.
Instead of building the runtime structure from zero, people start from a path that is already straight:
- identity has a home
- soul has a home
- work mode has a home
- customization starts immediately
That changed the feeling of the onboarding from “interesting, but a bit fuzzy” to “I can actually do this.”
And that is exactly what onboarding should do. Not impress people. Activate them.
The Shift In The Room Was Obvious
The two developer reactions were almost boring in the best possible way: “wow, with this method we can really be much faster.”
That is what you would expect.
What mattered more was the PO side.
By Friday, both had first PRs. Not because they turned into developers in a week. Not because engineering suddenly became unnecessary.
But because the distance between intent and first implementation artifact collapsed.
That changes the conversation completely.
A PO with this setup can move from:
- abstract requirement
- to executable first version
- to reviewable pull request
much earlier than before.
The developer no longer starts from zero. The developer starts from something concrete that already carries domain intent.
That is not replacement. That is compression of the translation layer.
This Is The Real Organizational Lever
People still talk about AI as if the main question were whether developers can code faster.
Of course they can. That part is obvious already.
The more structural question is this:
what happens when the people closest to the business problem can produce better first artifacts themselves?
Then a few things shift at once:
- handoffs get shorter
- misunderstandings surface earlier
- developers spend less time translating and more time judging
- domain understanding enters the artifact sooner
That is why I care about POs opening PRs.
Not because everyone should now cosplay as a software engineer. But because it proves that the system lowered the barrier between knowledge and execution without lowering the quality bar.
That is the kind of shift that compounds.
What I Would Measure From Here
If I were rolling this out more broadly, I would not start with vanity metrics like “hours saved” or “prompts used.”
I would track:
- who reaches first real artifact in week one
- who reaches first PR in week one
- how much review correction is needed
- whether domain clarity enters earlier than before
Because those are operational signals. They tell you whether the workflow changed, not just whether people had fun in a workshop.
And yes, the emotional side matters too.
Pride matters. Ownership matters. The sentence “I know now what to do” matters.
But it matters most when it is attached to a real output.
The Point
The best outcome of AI onboarding is not that people say “this is impressive.”
The best outcome is that people who used to stop at intent now move one step deeper into execution with confidence, structure, and visible results.
That is what happened in this workshop.
2 developers saw speed. 2 POs shipped first PRs. And the new template folder made the path into the system much straighter than before.
That is not hype. That is a workflow change.