AI-assisted delivery, from a manager's chair
How a product owner's job changes when AI does the building — your leverage moves to the brief, the guardrails, and the bar for 'good'.
By Renaud Yasin
- AI
- Claude
- Delivery
- DevSecOps
The hype around AI-assisted development is loud. From where I sit — owning the brief and steering delivery, and increasingly building through Claude myself — the real value is quieter and more durable than the demos suggest. It shows up in the unglamorous work, and it shows up in a less obvious place: what my own job becomes once the AI is doing the typing.
Where it earns its keep
For the teams I steer, AI tools like Claude pay off most in three places:
- Code review. Not replacing reviewers — giving them a faster first pass, so human attention goes to the judgment calls instead of the boilerplate.
- Documentation. The thing everyone deprioritizes. AI lowers the cost of writing it down enough that it actually gets written.
- Operational workflows. Runbooks, incident summaries, the glue work of 24×7 support. Small time savings, repeated constantly, add up.
But the speedups aren't the part that changed how I work. When AI does the building, a product owner's leverage moves up the stack — to setting the brief, drawing the guardrails, and deciding what "good" has to mean. Claude will clear whatever bar you set; choosing the right bar, and refusing to lower it when a looser one would flatter the result, is the actual work. The scarce thing was never the code. It's the judgment about what's worth building and what "done" honestly means.
What doesn't change
The fundamentals hold. You still need clear ownership, disciplined change management, and someone accountable when the pager goes off. AI makes a good delivery system faster; it makes a chaotic one fail faster. The operating model matters more than the tool — and the person setting the brief matters more than ever, because the AI builds exactly what you ask for, not what you meant.
How I'm using it
Across the engagement, we fold AI into the existing lifecycle rather than bolting on a separate "AI process." It lives inside code review, inside the docs, inside the ops workflow — where the work already happens. And on my own projects I take it further: I set the direction and the guardrails and build the thing through Claude end to end, then review every output before it ships. That shift — from writing the work to owning the brief and the bar — is the part I'll keep writing about here.