Anthropic's "Code with Claude" Event Unveils a Future Where AI Writes the Code

The atmosphere was electric at "Code with Claude," Anthropic’s two-day developer event held in London, which commenced on May 19th, coinciding with Google’s I/O. A startling moment occurred when Jeremy Hadfield, an Anthropic engineer, asked the packed room how many had shipped a pull request entirely written by Claude in the past week. Nearly half the attendees, many actively coding on their laptops, raised their hands. Even more remarkably, when asked who had done so without reading the code at all, most hands remained aloft, met with nervous laughter. Pull requests, fundamental to software development, represent fixes or updates submitted for review before deployment. Traditionally, these code chunks are the core work of professional developers. However, the advent of large language model (LLM)-powered tools like Anthropic’s Claude Code and OpenAI’s Codex has profoundly reshaped this process. Top tech companies now openly boast about the minimal amount of code their developers write by hand, with Anthropic stating that "most software at Anthropic is now written by Claude," and even "Claude has written most of the code in Claude Code." This rapid adoption highlights how quickly this new paradigm has become normalized. Anthropic’s ambitious goal is to push automation to its absolute limit, moving beyond AI-generated code requiring human cleanup. Their vision, articulated by Boris Cherny, head of Claude Code, is for Claude to "prompt itself" and autonomously check and correct its own work. This means human developers ideally wouldn't even see error messages; Claude would continuously test and tweak until the code functions perfectly. A new feature, "dreaming," announced recently, allows Claude Code agents to write and consolidate notes about tasks, enabling future agents to learn from past errors and improve their understanding of a codebase, essentially allowing Claude to "learn" and "get better." While "Code with Claude" radiated enthusiasm, featuring product showcases and workshops from companies like Spotify and Delivery Hero that have restructured their development teams around Claude Code, a different sentiment brews outside the conference halls. Online forums like Reddit and Hacker News reveal growing unease among coders. Many complain that AI coding tools, pushed by managers seeking productivity gains, paradoxically complicate development by increasing the volume of AI-generated code that still requires human review. Concerns also include a perceived decline in human coding abilities as tasks are offloaded to AI, and warnings from researchers about AI-generated code potentially introducing security vulnerabilities. Addressing these concerns, Katelyn Lesse, Claude engineering lead, and Angela Jiang, Claude product lead, emphasized that "all of the old software development best practices still apply." However, they acknowledged the increasing temptation to offload more tasks, including oversight, as automation advances. Lesse noted that some technical managers at Anthropic are already overwhelmed by the sheer volume of code their teams now produce. While Claude is currently considered as capable as a mid-level engineer, requiring expert human oversight for system design and complex troubleshooting, the ultimate goal is for Claude to "get better and better at all different types of engineering," with Jiang envisioning an "absolute end state" where "Claude basically being able to build itself." This rapid AI advancement, as noted by Stanford's 2026 AI Index, suggests humanity is struggling to keep pace.