# AI Coding
← All tags · 2 post(s)
16 Senior Devs Used AI to Code. They Thought It Made Them 20% Faster. It Made Them 19% Slower.
In METR's randomized controlled trial, 16 experienced open-source developers did real tasks on projects they'd maintained for an average of five years. The ones using AI were 19% slower. But the worse part is the other half: they predicted AI would speed them up 24% beforehand, and after finishing — after personally living through the slowdown — they still believed they'd gone 20% faster. Their gut and the stopwatch were off by nearly 40 percentage points, with the sign flipped. As someone who plans roadmaps, quotes timelines, and defends budgets on team-productivity estimates every day, I want to spell out where this illusion comes from, where it holds, and how it's quietly seeped into every AI-related decision in our line of work.
AI Coding Isn't Too Expensive — Nobody's Measured What It's Worth
Microsoft quietly pulled Claude Code from an internal division and pushed thousands of engineers back to GitHub Copilot. Uber burned through its entire 2026 AI coding budget in four months. The narrative is that AI coding is too expensive. It isn't. The real problem is that companies bought 'productivity gains' as a feeling, never as a number — and now the bill is crystal clear while the benefits aren't worth a single data point.