All posts
September 10, 2025·6 min read
Leading Claude AI adoption on a senior engineering team
What worked, what didn't, and the productivity gains we actually measured after rolling out Claude across the team.
AIClaudeTeam
Start with a real problem
When I introduced Claude to the team, I didn't pitch it as "AI will make us 10x." I pitched it as: _"Our code review backlog is 3 days deep. Let's see if Claude can cut that."_
A real, measurable problem. Not "innovation."
What worked
- Claude Code for stubs, scaffolding, and refactors: huge time saver, low risk.
- Claude for first-pass code reviews: catches obvious bugs, frees humans to think about design.
- Treating Claude like a junior: give it context, review its output, never merge blind.
What didn't
- Anything involving novel business logic. The model hallucinates plausible-looking code that's subtly wrong. Always verify.
- Asking for "an opinion" on architecture without constraints. Outputs are too generic to be useful.
The measurable result
Median PR-to-merge time dropped from ~36h to ~14h over six weeks. We didn't ship faster because Claude wrote our code. We shipped faster because Claude removed friction from review and refactoring.
The best AI workflow is the one your team forgets they're using.