Three months into Claudee Code](/tools/claude-code/)'s commercial availability, Anthropic held a two-day developer conference in London. Code with Claude kicked off on May 19, the same day Google staged its I/O developer event in Palo Alto. Anthropic staffers said the scheduling overlap was coincidental.
One of the opening questions to the audience: "Who here has shipped a pull request in the last week that was completely written by AI?" A pull request is the standard unit of code delivery - a bundle of changes submitted for team review before it goes into a live product. It's not a draft. It ships.
That question working as a casual icebreaker tells you where professional software development is in 2026.
The Normalization Happened Faster Than Expected
A year ago, the frame for AI coding tools was productivity: developers used them to autocomplete, write boilerplate, and speed up tasks they already knew how to do. The framing at Code with Claude was different. Claude Code, Cursor, and Aider have been in production long enough that AI-written pull requests are routine for a significant share of working developers. The question isn't whether to use AI for coding. It's how much of the codebase is AI-generated and how to manage that.
The skills shift is real. A developer who knows how to direct an AI agent, catch its failure modes, and review its output isn't doing the same job as one who writes code from scratch. Both are valuable. They're not the same.
What Running Parallel to Google I/O Signals
Regardless of Anthropic's stated intent, running a developer conference opposite Google I/O has an effect. Both companies are competing directly for developer attention. Google has Geminii Code Assist](/tools/gemini-code-assist/) and deep roots with developers through Android, Cloud, and open-source tooling. Anthropic has Claude Code and a model that developer communities have rated highly for coding tasks.
Code with Claude wasn't a community gathering. It was a statement about where Anthropic sees Claude Code relative to its competition - and the event drew enough developers to London to make that statement credible.
The Questions That Weren't on the Agenda
The harder issue for developers wasn't explicitly addressed but is clearly on many minds: what does code ownership mean when a model wrote it?
Code review has traditionally served multiple purposes - quality control, yes, but also teaching, knowledge transfer, and collective understanding of how a codebase works. When a senior engineer reviews a Claude Code-generated pull request, they're auditing work rather than mentoring the person who wrote it. Junior developers learning by reviewing AI output develop different instincts than those who learn by writing and getting detailed feedback.
These aren't reasons to abstain from AI coding tools. The teams ahead of the curve aren't abstaining. They're navigating this without a clear playbook, and that's the honest state of professional software development right now.