Seed-stage product team — AI workflow coaching
⚠️ This is an anonymised scenario narrative. It illustrates the shape of a coaching engagement. It is not a real client story and contains no real names, metrics, or claims about a specific company.
The problem
The team was small, fast, and using AI in the editor every day. Velocity was high. Two concerns were starting to surface:
- Code review had become a bottleneck. Reviewers were wary of approving AI-heavy PRs without a way to verify what the assistant had done.
- Junior engineers were learning to prompt, not learning to debug. The founder wanted to flip the ratio.
What we did
Workshop week
- Half-day one — Prompt engineering for code. Live exercises on a stripped-down sample repo. Patterns that transfer: from the vague request to the one with acceptance criteria.
- Half-day two — Quality and security gates for AI output. What to put in pre-commit, what to put in the PR template, what to put in CI. Live exercises on dependency drift, secret scanning, and “does this code even run”.
Both workshops were recorded and added to the team’s internal wiki.
Coaching month (4 weeks)
One 60-minute session per week with the two tech leads. Async questions from the rest of the team, answered within one working day. Topics were anchored on what they were shipping that week — never abstract.
The result (qualitative)
- The PR template gained a short “what I changed and why” section. Reviewers reported back that reviews felt faster and less subjective.
- Juniors spent more time debugging their own prompts and less time re-running them.
- The founder stopped reading every PR personally.
- The team kept the recordings and replayed them when onboarding a new hire a few weeks later.
[See the Team Coaching & Training page for the format that produced this engagement.]