At incident.io, nearly 80% of pull requests headed for production now include AI-generated code, according to data the software reliability company shared with The Pragmatic Engineer. The same account says experienced engineers increased their output, while newer engineers did better when they abstained from AI during their first month.
Watch Desk analysis
What happened
The company shared data on its AI use, code generation and productivity over the past two years. Its monthly spending on AI tools peaked during the summer, then fell significantly after it adopted newer models, according to The Pragmatic Engineer’s account.
The results point to different outcomes by experience level: tenured engineers increased output after using AI tools, while newer engineers performed better when they avoided them in their first month. The supplied account does not give the underlying measurements or define how output and performance were assessed.
Why it matters
This is a useful counterweight to the idea that one AI policy will suit every developer. The company’s account suggests that both the economics and the effect on work can change as tools and employees gain experience. The headline adoption figure is striking, but it does not tell us whether AI-written code is better, faster to review or more reliable in production.
The junior-engineer finding is the detail managers may find hardest to wave away. If early-career developers benefit from learning without AI for a time, blanket adoption targets could miss what helps them build skills. That is a possibility raised by this account, not a proven rule for every team.
Our read
The most useful takeaway is not “80% means AI has won”. It is that adoption, spending and productivity are moving together in less tidy ways than a simple usage chart suggests. Teams should measure quality and review effort alongside how much code AI contributes, and should test whether new starters need a different approach.
What to watch
- Whether incident.io shares the methods and measures behind its productivity findings.
- Whether newer engineers’ first-month results hold up across different teams and tasks.
- How spending changes as teams adopt newer models and assess the code they produce.
Discussion spark: Should software teams encourage new engineers to work without AI during their first month, or would that delay useful experience with the tools they will actually use?
Sources and evidence
- The Pulse: Details of incident.io’s AI spend (12 November 2026, 19:00 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.