Anthropic Engineer Says Claude Code Now Writes 100% of Her Team's Code

Inside Anthropic's All-AI Engineering Team

Anthropic's platform engineering group — the team responsible for the company's core services, developer tools and product infrastructure — no longer writes its own code. Katelyn Lesse, who leads the roughly 200-person group, said at Grafana Labs' ObservabilityCON that Claude Code, Anthropic's AI coding tool, now produces every line of code the team ships. "All of our code is written by Claude," she said. "Like, literally all of it."

The human role has not disappeared, but it has moved up the stack. Lesse said her engineers still design systems and build verification processes, then hand the actual coding to AI agents. When the team migrated a major codebase from Python to Rust, staff focused on defining how to verify the final product, while agents handled what she called the "grunt work" of building and monitoring the pieces. Anthropic already lets agents fix tests autonomously, and Lesse said the company is weighing how to let agents make changes to the live codebase directly — a step she acknowledged makes engineering leaders nervous.

The claim is part of a broader pattern at Anthropic. In May, the company said Claude wrote more than 80% of its entire production code, and a report last week noted that as AI agents took on more internal work, the company hired a smaller share of junior technical staff. Lesse framed the shift as a way to stay lean: Anthropic is approaching an IPO that could value it at nearly $2 trillion, even as its revenue growth slowed in the third quarter of this year.

What 'All Code Written by Claude' Really Means for Anthropic and Its Rivals

Why Anthropic Is Making This Argument Publicly

Lesse's remarks are not just an engineering update — they are a sales pitch. Anthropic's valuation depends on convincing enterprises that its AI-first model is replicable and worth paying for. If a 200-person team can run core infrastructure with agents writing all the code, the company can argue that Claude Code delivers the kind of productivity gain that justifies premium pricing and broad adoption. The timing matters: revenue growth slowed in the third quarter, so the IPO narrative needs a credible efficiency story to support a near-$2 trillion valuation.

The Junior Hiring Signal

Anthropic's own report that it hired a smaller share of junior technical roles as agents took on more work is the most concrete labor-market data point in the story. It suggests the productivity gains are real enough to change headcount composition, not just task allocation. For the broader software industry, this is the early evidence of a shift that could compress entry-level engineering demand while raising the premium on system design, verification and oversight skills — the tasks Lesse says humans still own.

What 'All Code Written by Claude' Does Not Tell Us

The claim is striking but narrow. It covers one team inside the company that builds the AI tool in question, which creates an obvious incentive to demonstrate the tool's capabilities. The article does not specify defect rates, incident frequency, security review outcomes or how much human time goes into verification — all of which determine whether the model is genuinely more efficient or simply relocates effort. Anthropic's own caution about letting agents change the live codebase autonomously suggests the company itself sees unresolved risk in the approach.

Where This Leaves Anthropic's Competitors

If Anthropic can credibly show that its internal engineering is agent-driven, it gains a reference case that rivals like OpenAI and Google must match or counter. The competitive pressure runs in two directions: Anthropic needs external customers to validate the model, while competitors can point to their own internal AI adoption as proof they are not falling behind. For enterprise buyers, the practical question is whether the productivity gains hold outside Anthropic's unusually AI-native environment, where engineers are building the very tool they use.

What Engineering Leaders Should Take From Anthropic's Experiment

Anthropic's claim is a strong signal about where AI coding is heading, but it is also a vendor making a case for its own product — treat the productivity numbers as directional, not settled.

• For engineering leaders: The tasks Lesse says humans still own — system design, verification, defining how to validate a migration — are the ones to invest in now. If your team is piloting AI coding tools, measure defect rates and review time, not just lines of code generated.
• For hiring and workforce planners: Anthropic's shift toward fewer junior technical hires is a concrete data point. Consider whether your entry-level pipeline needs to emphasize oversight, testing and architecture skills earlier.
• For enterprises evaluating Claude Code: Ask vendors for incident data and security review outcomes from their internal deployments, not just adoption percentages. The article notes Anthropic itself is still cautious about letting agents modify live code autonomously.
• For investors watching Anthropic's IPO: The near-$2 trillion valuation rests on enterprise adoption of this model. Watch whether Anthropic discloses customer productivity metrics or retention data alongside the efficiency narrative.