According to Fortune, Boris Cherny, head of Anthropic’s Claude Code, hasn’t written any code himself in over two months, with 100% of his code now generated by AI. He shipped 22 pull requests in one day and 27 the day before, all written by Claude. Anthropic CEO Dario Amodei said at Davos that some engineers have stopped coding, predicting AI could handle most software engineering from start to finish in just six to twelve months. An OpenAI researcher using the pseudonym Roon also stated he writes “0%” of his own code now, declaring “programming always sucked.” Across Anthropic, Cherny claims “pretty much 100%” of code is AI-generated, a shift enabled by their own Claude Code tool, which they used to build a new file agent called Cowork in about a week and a half.
The Unshackled Engineer
Here’s the thing: the most striking part of this isn’t the technical capability. It’s the emotional response from the engineers building these systems. Cherny told Fortune, “I have never had this much joy day to day in my work,” because Claude handles the tedious stuff and he gets to be creative. The OpenAI researcher Roon was even more blunt, calling programming a “requisite pain” and saying he’s glad it’s over. That’s a massive shift in mindset. For decades, the act of writing code *was* the creative, defining act of software engineering. Now, at the frontier, it’s being framed as grunt work to be automated away so they can focus on higher-level design and product thinking. They feel, in Cherny’s words, “unshackled.”
Hiring Habits and the Junior Dev Problem
This is where the implications get real, fast. Cherny says his team now hires “mostly generalists rather than specialists,” because the AI can fill in the implementation details. That’s a huge change. The traditional path into tech—master a specific language or framework, grind through entry-level bug fixes and feature work—is being disrupted. If the AI writes the first draft of everything, what’s the training ground? Fortune notes that open roles for entry-level software engineers have declined as AI-generated code ramps up. It’s not definitively causal, but the timing is hard to ignore. Companies might argue this democratizes building, letting non-coders create with natural language. But it could also concentrate opportunity, favoring those with high-level system design skills over those learning the craft from the ground up.
Not Perfect, But Accelerating
Let’s be clear: the code isn’t perfect. As AI researcher Andrej Karpathy pointed out in the thread that sparked Cherny’s revelation, models can make subtle conceptual errors and over-complicate things. The engineer’s job becomes reviewing, editing, and directing—a kind of super-powered technical product management. But the confidence from these engineers is that the quality will only improve. They’re betting the entire workflow on it. And when Roon says “100%, I don’t write code anymore,” it signals this isn’t just an Anthropic story; it’s happening at OpenAI, too. The labs building the most advanced models are becoming their own biggest, most demanding customers.
What Comes After Coding?
So what’s next? Cherny believes most of the industry will see “similar stats in the coming months,” and then we’ll start seeing AI take over non-coding computer work too. Think about all the admin tasks, data wrangling, and system orchestration that surrounds writing software. That’s the next frontier. Basically, the job is morphing from “software engineer” to “AI director” or “product specifier.” The core skill is shifting from writing syntax to writing impeccable instructions and having flawless technical judgment. It’s a fascinating, unsettling, and rapid transformation. The people who built the tools are the first to stop using the old ones. The rest of the tech world is just trying to catch up to the new reality they’ve created.
