source: techcrunch ai: coders are refusing to work without ai — and that could come back to bite them
level: technical
in february 2026, ai research lab metr tried to repeat a 2025 study on ai coding productivity but could not find participants. developers refused to work without ai tools, even for a limited number of tasks. the original study had shown that while developers felt more productive with ai, it actually slowed them down due to time spent fixing errors and guiding the ai.
a later metr survey in may 2026 found that technical employees believed ai made them twice as valuable. however, recent events cast doubt on these self-perceptions. amazon shut down an internal token-tracking leaderboard after employees gamed it, and uber exhausted its 2026 ai budget in four months without measurable productivity gains. programmer james shore argued that faster code generation may lead to higher maintenance costs, calling it a trade of temporary speed for permanent burden.
other evidence points to ai increasing code maintenance issues. a viral tweet claimed companies spend 44% of tokens on fixing ai-generated bugs, and code rabbit found ai produced 1.7 times more problems than human code. singapore management university researchers warned of long-term maintenance costs from ai-generated code. suggested solutions include using ai agents for fixes, but cognition's devin is rated between junior and mid-level. researchers recommend that programmers learn ai's strengths and weaknesses, implement quality assurance systems, and keep humans in charge of architecture and security.
why it matters: overreliance on ai coding tools may lead to hidden productivity losses and higher long-term maintenance costs, challenging the perceived benefits for software development.
source: techcrunch ai: coders are refusing to work without ai — and that could come back to bite them