A product team went from three AI licenses to a delivery pipeline it tunes itself, in six months.
At a glance
- Client: Soldd
- Engagement: six months of AI-native delivery advisory, 2026
- Focus: the agent pipeline and how the team measures it
- Results: median PR cycle time from 125 hours to 1.3 hours; high-severity escaped defects from 11 to 2
Before
The team’s AI setup was three Cursor licenses and one entry in the retro. The tools were there. A shared way of working with them was not.
The work
Order. Without help, the team expected to build “twice as many agents, in the wrong order.” My part was a sense of what mattered next, and the judgment to say which of the team’s ideas not to build.
Measurement. I kept measurement on the agenda until it stuck. The team now tracks PR cycle time, escaped defects, deploy frequency and velocity.
Experience from the same walls. I run the same experiments in my own practice, often in the same week. Much of the guidance came from walls I had already hit.
Cadence. Each 80-minute session had an agenda framed as questions, sent a day ahead. Each one ended with a written summary, with owners and dates, the same day.
After
From a Jira ticket, the team produces a spec, a plan, tasks and an estimate with 95% agentic support. Gates check the plan against the spec on the way in. They check the branch against its acceptance criteria on the way out. Low-risk changes merge without a human, and a release agent ships to production.
The specs, plans and the agents’ own rules live in the repo, under version control. When an agent does the wrong thing, the team finds the problem and fixes the configuration. Review is now more about the product than the code.
Results
- Median PR cycle time: 125 hours in May, 1.3 hours in September
- High- and highest-severity escaped defects: 11 in April, 2 in September
- Deploy frequency: up 74% over the latest three months, against the three before
- Velocity: up 51%, like-for-like
In his words
“A course gives you a framework. We got someone running the same experiments the same week, who’d already hit the walls we were walking toward. Her insights and guidance saved us at least 3 months of work.”
— Peter Riegler, Product Manager, Soldd
The longer story, and what I took from it: What Graduation Looks Like.