The One Paragraph That Explains Why Everyone's AI Productivity Numbers Are Contradictory

Clive Thompson's NYT Magazine piece is the best thing written so far on what AI is actually doing to a profession — not because it resolves anything, but because it goes inside the daily texture of the work instead of stopping at the labour market headline.

The thing it gets genuinely right: the greenfield/brownfield distinction. The founder building from scratch in a San Francisco apartment and the AWS engineer maintaining billion-line legacy systems are both called "software developers," but AI affects them so differently they might as well work in different industries. That's why you keep seeing "100x productivity" and "10% productivity" cited as contradictory facts. They're not. They apply to structurally different work. Thompson is the first writer I've seen actually name this and use it to explain the discrepancy, rather than picking the number that fits their argument.

The thing it doesn't earn: the skill atrophy thread. Two-thirds of the way through, a developer named Pia Torain mentions that four months of heavy AI use was enough to measurably erode her ability to code. That's the most alarming data point in the piece. It gets two paragraphs, then gets rebutted by a veteran who says Python didn't ruin anyone either. But the Python comparison doesn't hold — learning Python still required understanding computational thinking, and Python output was code you could read. The developer elsewhere in the piece who admits he "understands about half the work he produces" is describing something categorically different, and Thompson notices it, credits it, and moves on far too quickly.

The ending — abstraction may be coming for us all — is earned. The middle buries its own best finding.

For non-technical readers, the paragraph about Maxime Cuisy, the Paris print shop manager who built working software in an afternoon with no coding background, is the one to read twice. It's not a coding story. It's a story about what happens when the gap between "I have a problem" and "I have a tool that solves it" closes — and that gap is moving in your field too.