The engineers who stopped coding

Boris Cherny hasn't written a line of code in two months. As head of Claude Code at Anthropic, he ships 22 pull requests one day, 27 the next. All written by Claude.

Across Anthropic, engineers report 70-90% of their code is AI-generated. At OpenAI, researcher Roon says the same: "100%, I don't write code anymore." Then adds: "Programming always sucked. It was a requisite pain for everyone who wanted to manipulate computers into doing useful things, and I'm glad it's over."

The sentiment cuts deeper than productivity stats.

What programming actually was

Roon's comment reveals something uncomfortable. Programming wasn't the creative problem-solving we told ourselves it was. It was mostly typing out the obvious solution to a problem you'd already solved in your head.

The creative part happened before you touched the keyboard. You figured out the approach, sketched the architecture, identified the edge cases. Then came the tedious translation into syntax the computer could understand. Variable declarations, error handling, boilerplate. The programming equivalent of copying your essay in neat handwriting.

Cherny describes his current work as pure creativity: "All the tedious work, Claude does it, and I get to be creative. I get to think about what I want to build next." He's not writing code anymore. He's designing systems.

The specialization trap

Anthropic now hires generalists instead of specialists. "Not all of the things people learned in the past translate to coding with LLMs," Cherny notes. "The model can fill in the details."

This hits specialists hardest. The engineer who spent years mastering React performance optimization or database indexing strategies finds their expertise compressed into a prompt. The model knows the patterns. It handles the details. The specialist becomes a generalist overnight.

Meanwhile, domain experts who never learned to code suddenly can. They know what needs building. They understand the business logic. They just needed someone to translate their requirements into code. Now the AI does that translation.

The adoption gap

Outside AI labs, the numbers look different. Microsoft reports 30% AI-generated code. Salesforce similar. GitHub studies show 29% of Python functions in the US are AI-written.

The gap isn't technical. It's cultural. Anthropic engineers embraced the shift because they built the tools. They trust them. They've seen the quality improve weekly. Most companies are still testing, evaluating, waiting for approval from legal.

Cherny predicts the industry will catch up within months. "We will then start seeing similar stats for other computer work also." The pattern spreads beyond programming to any work that involves translating human intent into computer instructions.

What remains human

The engineers who stopped coding didn't stop building software. They stopped implementing it. Cherny still ships features, fixes bugs, optimizes performance. But he does it by describing what he wants, not by typing out how to do it.

This distinction matters. Knowing what to build requires domain knowledge, user empathy, system thinking. Knowing how to build it was always the mechanical part. The part computers excel at once they understand the patterns.

The shift forces a question: if you remove the coding from software engineering, what's left? The answer might be what the job was supposed to be all along. Understanding problems, designing solutions, making tradeoffs. The computer handles the typing.