AI Coding Is More Addictive Than Helpful Say Developers
Via ZDNET
The Story
Four in five developers say AI coding tools have come to feel more like a dependence than an advantage, according to a Coddy Tech survey of 305 programmers.
The same survey found 43 percent keep coding with AI after hours when they meant to stop, 32 percent put off sleep to continue, and 39 percent say the tools make it harder to switch off from work.
Quentin Rousseau, chief technology officer at the incident response company Rootly, described watching an agent refactor a module at 2:47 in the morning and being unable to stop, saying that success delivers a dopamine hit and failure an adrenaline rush.
Developers still see career value in the tools, with 74 percent expecting heavy use to improve their odds of a raise or promotion, though 51 percent also expect to burn out.
The 2025 Stack Overflow Developer Survey put adoption at 80 percent while trust in accuracy fell from 40 percent to 29 percent and favorability slipped from 72 percent to 60 percent.
Nearly half of those respondents were frustrated by answers that were almost right, which produce convincing output and difficult debugging, a burden now described as verification debt.
Teams that use agents to remove routine work may gain real time, while teams that accelerate every stage risk a faster and more relentless version of the same job, leaving developers to decide when the agent loop ends.
The same survey found 43 percent keep coding with AI after hours when they meant to stop, 32 percent put off sleep to continue, and 39 percent say the tools make it harder to switch off from work.
Quentin Rousseau, chief technology officer at the incident response company Rootly, described watching an agent refactor a module at 2:47 in the morning and being unable to stop, saying that success delivers a dopamine hit and failure an adrenaline rush.
Developers still see career value in the tools, with 74 percent expecting heavy use to improve their odds of a raise or promotion, though 51 percent also expect to burn out.
The 2025 Stack Overflow Developer Survey put adoption at 80 percent while trust in accuracy fell from 40 percent to 29 percent and favorability slipped from 72 percent to 60 percent.
Nearly half of those respondents were frustrated by answers that were almost right, which produce convincing output and difficult debugging, a burden now described as verification debt.
Teams that use agents to remove routine work may gain real time, while teams that accelerate every stage risk a faster and more relentless version of the same job, leaving developers to decide when the agent loop ends.
Go Deeper
Read the original reporting at ZDNET.
Read Full Story at ZDNET →