70% of developers say AI coding tools make them more productive. Only 17% say those tools improve team collaboration.
That stat from Stack Overflow's 2025 survey stuck with me. We're all shipping faster individually, but nobody's talking about the team-level side effects.
The numbers that should make you uncomfortable
A few data points that changed how I think about this:
- METR study: Experienced devs using AI tools took 19% longer to complete tasks, while believing they were 20% faster. A 40-point perception gap.
- DORA 2024: A 25% increase in AI usage correlates with a 7.2% decrease in delivery stability.
- GitClear: Code churn jumped from 3.1% to 7.9% between 2020-2024. Refactoring dropped from 25% to under 10%.
More code is shipping. Whether it's the right code is a different question.
The collaboration problem nobody's fixing
A two-year longitudinal study found that AI adoption shifts work toward individualized coding tasks and away from collaborative coordination. The collaboration problems that existed before AI (silos, communication gaps, unclear ownership) stayed completely unresolved.
AI is making individuals faster. It is not making teams better.
Where retros come in
When your team spends 90% of the sprint heads-down with an AI pair programmer, the 60 minutes in a retro might be the most important hour in the entire sprint.
The questions need updating though. For AI-assisted teams, these are the ones that matter:
- AI tool effectiveness - Where did AI help? Where did it waste your time?
- Knowledge distribution - Who actually understands the code that shipped?
- Customer connection - Did shipping faster translate into customer value?
- Code quality signals - Is churn going up? Are PRs getting rubber-stamped?
- Team norms - What are your unspoken rules about AI usage?
The DORA quote that sums it up
"AI makes good teams great. And bad teams worse, faster."
-- Google DORA 2025 Report
The practices that separate good teams from bad ones (psychological safety, shared understanding, honest feedback) are exactly what retros are built around. As AI handles more of the mechanical work, the human conversations get rarer. And rarer means more valuable.
I wrote a longer piece diving into the research and practical retro questions for AI-assisted teams: Read the full post on Kollabe
Top comments (3)
That 40-point perception gap from METR is the stat I keep coming back to as well. The code churn data from GitClear makes it concrete — we're not just slower, we're shipping work that gets re-done more often. The retro angle is smart because it catches the team-level effects that individual productivity metrics miss entirely. One thing I'd add: the retros need to explicitly ask "what did AI-generated code cost us this sprint" — not as a blame exercise but as calibration. Most teams I've talked to don't track this at all, so the 7.2% stability hit just accumulates invisibly until something big breaks.
Quite correct. As development moves at quicker paces, our team needs to pause, discuss and adjust our workflow. Our weekly meetings on this are also a way to exchange experiences and findings.
good job