A senior Google engineer has said that Anthropic’s Claude Code produced, in about an hour, a distributed agent orchestration system comparable to one her team at Google had spent the past year building. Jaana Dogan, a principal engineer who works on Google’s Gemini API, described the result in a post on social media in early January 2026, saying the tool worked from a problem description of only three paragraphs.
Public praise from a Google engineer for a rival company’s developer tool is unusual, and the post drew wide attention among software developers. Dogan noted a relevant caveat herself: Google is a major investor in Anthropic. Google has put roughly $3 billion into the company, and in October 2025 the two announced an expanded deal under which Anthropic will gain access to up to one million of Google’s tensor processing units, an agreement valued at tens of billions of dollars.
What Claude Code produced
Agent orchestration systems coordinate multiple AI agents that split up a task, call tools and pass results between one another, often across many machines. Building one that is reliable at scale involves scheduling, state management and failure handling, which is why it can occupy an engineering team for months. According to Dogan’s account, as reported by The Decoder and other outlets, Claude Code generated a working version of such a system in a single session after being given a short written specification.
Her broader point was about the pace of change in AI-assisted programming. In 2022, code assistants mostly completed a few lines at a time. By 2025, agentic tools such as Claude Code, which runs in a developer’s terminal and can read, edit and run code across an entire repository, were producing complete codebases and nontrivial system architectures.
The industry picture
Dogan’s observation fits with figures the large AI companies have offered about their own engineering. Anthropic chief executive Dario Amodei has said that the large majority of code at Anthropic, which he has put at around 90 percent in some teams, is now written by AI. Google chief executive Sundar Pichai has said that roughly 30 percent of new code at Google is generated by AI and then reviewed by engineers.
These numbers come from the companies themselves and are not independently audited, and a one-hour demonstration is not the same as a production system that has been hardened over a year. Even so, the shift is changing how engineering organizations think about team size, cost and the role of human programmers, who increasingly spend their time specifying problems and reviewing generated code rather than writing it line by line.
The competition among AI vendors is no longer only about which model scores highest on benchmarks. It is increasingly about how quickly a company, and its customers, can turn a written description into working software.
