Anthropic’s AI, Claude Code, now generates nearly all internal code, prompting questions about its 100+ open engineering roles. Executives clarify human engineers are crucial for prompting, customer interaction, and strategic decisions. This shift elevates engineers to focus on system design and product direction, with AI handling repetitive coding tasks. The trend signifies a significant evolution in software development.

Anthropic’s own executives have confirmed it—Claude Code is now writing nearly 100% of the company’s internal code. So when the company’s job page still shows 100+ open engineering positions, the obvious question is: what exactly are these engineers supposed to do Boris Cherny, head of Claude Code, addressed this directly when an X user flagged the apparent contradiction. His answer was simple. “Someone has to prompt the Claudes, talk to customers, coordinate with other teams, decide what to build next,” he wrote. “Engineering is changing and great engineers are more important than ever.”

Claude is writing Claude—but humans are still running the show

The backdrop here matters. At the 2026 Cisco AI Summit, Anthropic CPO Mike Krieger confirmed that the company’s AI tools now generate nearly all of its internal code. “Claude products and Claude code are being entirely written by Claude,” he said—building on CEO Dario Amodei’s earlier prediction that AI would handle 90% of coding, a figure that has since been surpassed inside Anthropic itself.Boris Cherny added more colour in a Fortune interview, saying he personally hasn’t written a single line of code in over two months, shipping 22 to 27 pull requests a day—all 100% Claude-generated.

AI is handling the grunt work—engineers are making the bigger calls

What this means in practice is that engineers have moved up the stack. Repetitive tasks—drafting features, refactoring, writing documentation—now belong to Claude. Humans focus on system design, architectural decisions, customer conversations, and deciding what gets built next. Everything AI produces is still reviewed and validated before it ships.The trend isn’t unique to Anthropic. Microsoft and Salesforce report around 30% AI-generated code, while a recent Science journal study put the figure at roughly 29% for GitHub Python functions in the US. Inside the frontier labs, though, the curve is far steeper—and Cherny thinks the rest of the industry will get there soon.