Code Claude: A Paradigm Shift in Coding Approaches

AI is paving the way for the future of coding. Anthropic held its developer event, 'Code with Claude,' in London, unveiling an era where AI writes code. The event, held over two days starting May 19th, featured discussions on how AI can innovate development tasks.

On stage, Anthropic engineer Jeremy Hadfield asked, "How many people submitted a pull request entirely written by Claude last week?" Nearly half the attendees, who were coding or typing prompts on their laptops while watching the presentation, raised their hands. A pull request is a core component of code written by professional developers, representing a proposed modification or update to existing software, which is then subject to review. Hadfield followed up, "How many submitted a pull request written by Claude without reading the code?" A hesitant chuckle rippled through the audience, and most hands remained raised.

AI-powered tools like Claude Code are transforming how software is created. "Most of Anthropic's software is now written by Claude," stated Anthropic's Hadfield. OpenAI, Google, and Microsoft make similar claims. Anthropic has also held developer events in San Francisco and Tokyo. Around this time last year, the company had just launched Claude 4. While it possessed coding capabilities, Anthropic's latest updates, particularly Claude 4.6 and 4.7 released in February and April, have made Claude Code a tool that developers are increasingly willing to entrust with their work.

Anthropic aims to push automation to its limits. Instead of humans editing and correcting errors in AI-generated code, they want Claude to verify and correct its own work. Boris Cherny, head of Claude Code, explained, "The default is not to give Claude a prompt, but to have Claude prompt itself." Claude handles everything, going through testing and refinement until it functions correctly. "The core principle is to let Claude work uninterrupted. We say, 'Let's let this work,'" explained Anthropic engineer Ravi Trivedi in another talk.

Anthropic introduced a new feature called 'Dream.' The Claude Code agent writes notes to itself, recording and storing useful information about a specific task. Later, when another coding agent works on the same code, it can use these notes to get up to speed faster and learn from errors made by previous agents. Dream helps Claude Code learn about a specific codebase and improve its work capabilities. This is expected to impact the domestic software industry as well.