AI Coding Speeds Implementation but Raises Engineering Burden, as 200-Person Study Tracks Higher Expectations
Updated
Updated · InfoWorld · Sep 29
AI Coding Speeds Implementation but Raises Engineering Burden, as 200-Person Study Tracks Higher Expectations
3 articles · Updated · InfoWorld · Sep 29
Summary
AI coding agents can accelerate implementation without making engineering easier, because the saved time often shifts developers toward harder decisions on review, compatibility, maintenance and scope.
Simon Willison argued the tools demand “extraordinary discipline and knowledge,” while enthusiastic users said the work can become more taxing rather than lighter even when output rises.
An 8-month UC Berkeley study at a 200-person tech company found employees used AI to keep more work moving and expand responsibilities, with extra effort gradually becoming the new baseline.
That dynamic matters for budgets: companies may mistake a burst of AI-enabled output for durable labor savings when some gains come from longer hours or from taking on more ambitious projects.
Willison’s September Datasette security audit showed the trade-off—AI surfaced more vulnerabilities, but two human maintainers still had to test, verify and ship fixes across main and stable branches.