AI Is Writing More Code. Programmers Are More in Demand Than Ever.

The fear is reasonable on its face. If AI can write code, and it increasingly can, then surely the humans who write code will eventually find themselves surplus to requirements. Snap recently disclosed that AI writes 65% of its code. China's GLM-5.1 topped major coding benchmarks last week. GitHub Copilot, Cursor, and Claude handle substantial portions of routine coding tasks at companies of every size. The machines are getting good.

The employment data is not cooperating with this story.

The US Bureau of Labor Statistics projects software developer roles will grow well above the national average through 2033, a forecast that has not been meaningfully revised downward despite years of AI coding tool adoption. The companies building and deploying the most capable AI coding tools are, broadly, hiring more engineers than they are letting go.

Part of this is structural. When the cost of producing something drops, demand for it tends to rise rather than fall. Spreadsheets made financial analysis faster, but they didn't shrink the number of financial analysts. They expanded what analysis was used for. The same dynamic appears to be playing out in software. AI coding tools make it cheaper and faster to build software, and the response from businesses so far has been to build more of it.

Second, AI coding tools are amplifiers, not substitutes. GitHub's own research found that developers using Copilot complete coding tasks substantially faster, but the tool doesn't replace the engineering judgment about which tasks to do, how to structure a system, what tradeoffs to make, or how to diagnose a production failure that doesn't resemble anything in the training data. That judgment still requires people.

Third, there's the AI work itself. The infrastructure required to build, evaluate, and deploy AI systems is engineering work that barely existed at scale five years ago. A meaningful share of current developer demand is directly driven by the AI wave people assume is displacing engineers.

None of this means the profession is unchanged. The mix of skills that makes a developer valuable is shifting. Developers who can work effectively alongside AI tools, who know how to prompt, evaluate, and sanity-check generated code, are more productive than those who can't. Some entry-level work that involved writing boilerplate is being restructured. That's real.

But "the shape of the work is changing" and "programmers are obsolete" are very different claims. As we noted in earlier coverage of the job research, the industries most exposed to AI are mostly adding jobs. Software development is among the most AI-exposed industries in the economy.

The concern worth taking seriously isn't mass obsolescence. It's narrowing entry paths: AI-assisted productivity may raise the output bar for individual contributors, making it harder for junior developers to get hired and build the experience that leads to senior roles. That's a genuine structural question. It's just not the apocalypse the headlines keep gesturing at.