Software development jobs hanging on a thread as companies increasingly use AI to write code

Anthropic lead shocks industry by abandoning manual coding entirely; CEOs predict "end-to-end" automation within 12 months.

The traditional image of a software engineer hunched over a keyboard, meticulously crafting lines of logic, is rapidly becoming a relic of the past. In a series of bombshell revelations this week, top engineers at the world’s leading AI labs admitted they have almost entirely stopped writing code by hand, marking a "singularity moment" for the profession that has defined the digital age.

The shift is no longer theoretical. Boris Cherny, the head of Anthropic’s Claude Code, recently announced that he hasn’t written a single line of manual code in over two months. Instead, 100% of his output, including 49 pull requests (PRs) in just 48 hours, was generated by Claude Opus 4.5.

The shift is not limited to Cherny’s workflow. Across much of Anthropic, he estimates that nearly all code is AI-generated. He says that at Anthropic, "pretty much 100%" of code is now AI-generated, with human engineers shifting their focus entirely to high-level architecture and "editing" rather than implementation.

"Programming Always Sucked"

The sentiment was shared by Roon, a pseudonymous OpenAI researcher, who told his followers on X, "100%, I don’t write code anymore. Programming always sucked... I’m glad it’s over."

This internal reality at AI labs is far ahead of the broader market, but the gap is closing fast:

• Microsoft: CEO Satya Nadella reported in April 2025 that 30% of the company’s code is AI-written.

• Google: CEO Sundar Pichai has cited similar figures, noting that over a quarter of all new code is AI-generated.

• Industry Wide: A recent study in the journal Science found that 29% of all Python functions in the U.S. are now authored by AI.

The "Cowork" proof of concept

The speed of this transition was recently demonstrated by the launch of Cowork, Anthropic's new file-management agent for non-coders. Cherny revealed that his team built and shipped the entire product in just one and a half weeks.

The "force multiplier" effect was clear: Cowork was built almost exclusively using Claude Code, proving that AI can now be used to rapidly birth new AI products without the months-long development cycles typically required by human teams.

A reckoning for entry-level roles

While veterans like Cherny celebrate "creative freedom," the shift has created a precarious environment for those just entering the field. The traditional career path, starting as a junior dev and "learning the ropes" through routine coding tasks, is effectively being automated out of existence.

• Hiring shifts: Anthropic has already pivoted its hiring strategy, moving away from "specialists" (who know specific languages deeply) to "generalists" who can manage AI agents.

• Junior collapse: Open roles for entry-level engineers have declined sharply as companies realize a single senior engineer equipped with AI can match the output of a former 5-person junior team.

The "Skill Gap"

Still, AI researcher Andrej Karpathy warned that while AI is fast, it can still make "subtle conceptual errors." Without a new generation of humans learning the "guts" of coding, the industry risks a future leadership vacuum where no one truly understands the systems the AI is building.