325
Audio & Video Production313
Software Development239
Automation & Workflow204
Marketing & Growth185
AI Infrastructure & MLOps145
Writing & Content Creation190
Data & Analytics114
Design & Creative148
Photography & Imaging140
Customer Support119
Voice & Speech130
Sales & Outreach110
Education & Learning120
Operations & Admin80
New reports say AI coding tools help generate more code, but teams often spend more time revising it later, which can reduce real productivity.
In short: More developers are using AI coding tools to write faster, but new data suggests a lot of that code has to be rewritten later.
Companies are increasingly giving software developers large “token budgets.” Tokens are the units that many AI tools charge for, similar to paying by the number of words and steps the tool uses. Some developers treat a bigger token budget as a status symbol, but a TechCrunch report says that does not necessarily mean more useful work gets done.
Several analytics firms that track software work say AI tools can lead to more code being created, but also more “code churn.” Code churn means code that gets added and then later deleted or rewritten, like writing a long draft and then throwing much of it away.
Waydev, which works with 50 customers and more than 10,000 engineers, says managers may see AI code acceptance rates of 80% to 90% at first. But Waydev CEO Alex Circei told TechCrunch that follow-up revisions in later weeks can drive the real kept-and-working amount down to 10% to 30% of what was generated.
Other data points in the story point in the same direction. GitClear reported in January that regular AI users had 9.4 times higher churn than non-AI users. Faros AI reported in March 2026 that churn increased 861% under high AI adoption. Jellyfish found that engineers with the biggest token budgets produced more pull requests (proposed code changes), but the gains did not scale, with about two times the output at ten times the token cost.
More companies are likely to focus less on how much AI is used and more on whether the code “sticks” over time. Tools that measure quality, rework, and cost may become a bigger part of managing AI-assisted software work.
Source: TechCrunch AI