In 30 seconds
- AI speeds up building; it doesn't remove uncertainty about what to build.
- Agile's core loop (small steps, real feedback, adapt) matters more, not less.
- Practices built around human typing speed need rethinking.
What stays the same
The hard part of product work was never typing. It's knowing what users need, making good trade-offs and learning fast. The agile values and empiricism still apply: build a little, check it with real users, adapt.
What changes
| Area | Shift |
|---|---|
| Bottleneck | Moves from writing code to deciding, specifying and reviewing |
| Batch size | Can shrink further: cheaper to build and throw away |
| Estimation | Effort-based sizes lose meaning; see estimation with AI |
| Quality | Review and testing carry more weight; see human-in-the-loop review |
Where to start
- Decide where your team sits: AI-assisted or AI-native.
- Update the Definition of Done for AI-generated work.
- Watch quality signals like change failure rate as speed rises.
Myth vs reality
Myth · AI makes agile obsolete.
Reality · Faster building makes fast feedback more valuable.
Myth · AI just makes teams faster at the same work.
Reality · It changes where the bottleneck sits.
Myth · Adopting AI tools is an agile transformation.
Reality · Tools change little without new habits.
FAQ
Should we change our framework?
Usually not; adapt practices within it first.
What's the biggest risk?
Shipping more code faster than the team can review and understand it.
Who should lead the change?
The team, with leaders setting guardrails.