Your AI Coding Assistant Might Be a Paperweight in Disguise

A new study from the respected AI research lab METR reveals a startling shift in developer behavior: as of early 2026, most programmers refuse to code without AI assistance, even for short tasks. The lab attempted to replicate a 2025 experiment measuring AI’s impact on coding speed, but couldn’t find participants willing to work unassisted. That’s a red flag for engineering leaders betting big on AI productivity gains.

Why? Because the data suggests AI isn’t making code better—it’s making it messier. Developers in the original study thought AI boosted their output, but when timed, they were actually slower after accounting for error correction and tool steering. Now, new research from Singapore Management University warns that AI-generated code introduces long-term maintenance costs. Code review firm Code Rabbit found AI-produced pull requests create 1.7 times more problems than human-written code. Meanwhile, Amazon shuttered its internal token-tracking leaderboard after employees gamed it, and Uber burned through its 2026 AI budget in four months with no measurable productivity lift.

Programmer James Shore put it bluntly: “You write code twice as quick now? Better hope you’ve halved your maintenance costs. Otherwise, you’re screwed.” The takeaway for business leaders? AI coding tools are here to stay, but treating token volume as a productivity metric is a costly mistake. The smart play is to invest in robust QA systems, keep humans on architecture and security, and treat AI-generated code like work from a junior developer—review it carefully, don’t trust it blindly.