Coders are refusing to work without AI — and that could come back to bite them
Developers increasingly rely on AI coding tools, but evidence suggests the speed gains may bring higher error and maintenance costs.
Intelligence analysis by GPT-5.4 Mini
TechCrunch reports that many developers no longer want to work without AI, even as research and industry examples suggest the tools can slow some work, add bugs, and raise long-term maintenance costs. The piece argues that AI may boost output now while creating technical debt later.
Many coders like AI because it writes code fast, like a super-speedy helper. But fast does not always mean good. If the helper makes mistakes, people have to spend extra time fixing them.
The article says some studies found people felt faster with AI, but the real work sometimes took longer because they had to check and repair the code. It is like having a robot write a school essay very quickly, but then the student still has to rewrite half of it.
The main idea is simple: AI can help, but it should not be trusted blindly. People still need to watch it carefully, especially for important parts like safety and system design.
Analysis
The core tension
TechCrunch says AI coding tools have become so popular that some developers would not even join a study if they had to work without them. That finding came from METR, which had hoped to repeat earlier research on coding productivity.
What the research suggests
The earlier 2025 work found a mismatch between perception and outcome: developers felt faster with AI, but the measured results showed they were slower overall. The article says the slowdown came from the extra work of checking, correcting, and steering the model’s output. In METR’s follow-up effort, researchers could not run the same test because participants did not want to work without AI, so they switched to a self-reported survey in May. In that survey, technical workers said they believed AI made them about twice as valuable.
Why the optimism is being questioned
The story argues that confidence in AI productivity is running into cost and quality concerns. It points to “tokenmaxxing,” the practice of using token counts as a productivity proxy, and notes examples where companies had to pull back. TechCrunch cites a report that Amazon shut down an internal token leaderboard after employees gamed it with AI agents, and says Uber exhausted its 2026 AI budget within four months without a measurable productivity gain, according to reporting from The Information and comments from COO Andrew Macdonald.
The article also highlights maintenance risk. Programmer James Shore argued that faster code creation can still leave teams with higher ongoing costs if maintenance does not fall too. Aiswarya Sankar of Entelligence AI said companies are spending 44% of tokens on bug fixes caused by AI-generated code, while Code Rabbit says its analysis found AI code produced 1.7x more problems than human code. TechCrunch notes those figures come from vendors with a business interest, but says independent research from Singapore Management University reached a similar warning: AI-generated code can add long-term maintenance costs to real projects.
The recommended response
The piece says AI coding agents are not a full substitute for human engineers. Even Scott Wu, CEO of Cognition, described Devin as somewhere between junior and mid-level depending on the task. The article closes by saying programmers need to understand where AI is weak, keep strong quality assurance in place, and leave architecture and security design to humans.
Key points
- METR says many developers now refuse to work without AI, even for study participation.
- Earlier research found AI made developers feel more productive, but they actually finished tasks more slowly.
- The article warns that AI-generated code can create extra bugs and long-term maintenance costs.
- Examples from Amazon and Uber suggest heavy AI use does not automatically improve productivity or reduce costs.
- The piece argues humans still need to handle architecture, security design, and careful review.
