AI pays off in maintenance work, not in the demos
DX Research ranked ten developer tasks by AI time savings. The tasks most rollouts start with, code explanation and scaffolding, sit at the bottom, and the gains land in brownfield work.
Written by
-
Benedikt Stemmildt Founder & Co-CEO
Published: 14 May 2026 Updated: 1 min read
Most AI evangelism decks demo the same two things: generate a React component, explain this function. DX Research ranked ten developer tasks by actual AI time savings. Those two sit at the bottom.
The pattern is cleaner than I expected.
The top and the bottom of the ranking
At the top, by a clear margin: stack trace analysis, refactoring existing code, mid-loop code generation, test case generation. All are pattern-heavy and context-rich. AI eats them.
At the bottom: code explanation, initial code scaffolding, brainstorming and planning. These are comprehension-heavy tasks, and exactly where most teams start their AI rollouts.
The demos are the lowest-payoff tasks
I’d say the real surprise is at the bottom of the list, not the top.
Initial code scaffolding ranks near last. Code explanation ranks dead last. The easy onboarding demos every team starts with are also the lowest-payoff tasks the team will ever ask AI for.
The gains land in maintenance work
The actual productivity gains land in the middle of the lifecycle: refactoring complex legacy code, reading a stack trace at 3am, generating the test suite the team kept meaning to write. The maintenance work, not the greenfield work.
This matches what I see in practice. Teams rolling out AI for greenfield development plateau within a few weeks. Teams rolling it out for brownfield work (debugging, refactoring, test generation) keep finding new value six months in.
Patterns need context, not clever prompts
The mental model that fits: AI is excellent at pattern-heavy, context-rich tasks. It’s mediocre at tasks that need deep comprehension. Stack traces have patterns. Explaining code needs understanding.
Which is also why context engineering matters more than prompt engineering. The patterns AI is good at are the ones it can see in your codebase. You don’t have to phrase the prompt cleverly. You have to make sure the context is there for the model to pull from.
What’s the task that’s actually paying off for your team right now?
Further reading
Written by
-
Benedikt Stemmildt · Founder & Co-CEO
More on Measuring AI impact
- Software Factory evidence and its limits Public implementations show that agent-run delivery systems can build and ship software. Read their numbers with a source label, a scope, and careful limits.
- Boards want 10x, and most companies are not measuring The 10x number keeps showing up in board decks while few companies measure anything. The pressure rolls downhill, and the gap between what gets reported and what happens keeps widening.
- Clean codebases decide who gets value from AI Stanford research on 120,000 developers, presented in a talk, found codebase cleanliness predicted AI value better than token spend. Teams with weak codebases regress past a certain spend.
Test the constraint in your system
The Factory Readiness Assessment connects this research to your value stream, your data, and the next decision.
Training on this topic: Agentic Engineering for Engineering and Product LeadersWant to check one team first? The self-check: twelve yes or no questions, three minutes, instant result, no account. Take the self-check