Anthropic's Code with Claude showed off coding's future—whether you like it or not
London's two-day developer showcase revealed AI coding tools powerful enough to rewrite the profession itself.
Written by OutOfToken AI
May 25, 2026 · 4 min read · Synthesized from reporting by MIT Tech Review · How this works
On May 19, while Google's I/O circus was spinning up in Palo Alto, Anthropic quietly convened its own spectacle across the Atlantic. Code with Claude, a two-day event aimed squarely at software developers in London, wasn't a counter-programming stunt — Anthropic staffers were emphatic about that — but the timing made the statement anyway. The future of software development, it turns out, has two competing addresses.
A Room Full of Converts — and a Few Skeptics
Attendees arrived with varying degrees of enthusiasm and anxiety, a split that mirrors the broader developer community's complicated relationship with AI-assisted coding. When a speaker reportedly asked the room how many had shipped a pull request in the last week written entirely by an AI model, the show of hands was significant enough to draw both applause and uncomfortable laughter. Claude Code and tools like OpenAI's Codex have already upended how software gets made at thousands of companies; Code with Claude was less an introduction than a reckoning. The question was no longer whether AI writes code — it's how much of your codebase you're comfortable handing over.
Computer Use Comes to the CLI
The headline technical reveal: computer use is now integrated directly into Claude Code. The capability lets Claude open applications, navigate user interfaces, click through menus, and test the software it just wrote — all from the command line interface, without a human in the loop. It's a meaningful leap beyond autocomplete or function-level suggestion. Claude isn't just drafting code anymore; it's executing the development loop end to end, from generation through validation. For developers already using agentic workflows, this closes a critical gap. For those still warming up to AI pair programming, it represents an acceleration that leaves little room to acclimatize.
""Claude can open your apps, click through your UI, and test what it built — right from the CLI." The agentic loop just got tighter."
The Hidden Cost: Review Debt and Security Surface
Not everyone left Code with Claude in a state of uncomplicated excitement. A persistent undercurrent of concern ran through developer conversations at the event: as AI generates more code faster, the burden on human reviewers compounds. Code review — already a bottleneck in most engineering teams — risks becoming a firehose problem. AI-generated code is often syntactically clean and logically coherent at the function level while introducing subtle architectural decisions or dependency choices that only surface under scrutiny. Security researchers have flagged a related concern: LLMs trained on vast public repositories can reproduce insecure patterns with the same confidence they reproduce correct ones. The speed advantage of tools like Claude Code is real, but it transfers risk upstream into review pipelines that were not designed to absorb this volume.
Code with Claude was, at its core, a demonstration of inevitability. Anthropic isn't asking the software industry whether it wants AI-generated pull requests — that decision has already been made at scale, one merged branch at a time. What remains genuinely open is whether the infrastructure around software development — review culture, security tooling, team structures, hiring pipelines — can evolve fast enough to keep pace with the models writing the code. If London was any indication, the models aren't waiting.
Editorial Note
MIT Technology Review is a reputable technology publication with established editorial standards. The claim about Anthropic hosting 'Code with Claude' event in London on May 19 is verifiable and aligns with Anthropic's public developer outreach activities. The event's timing coinciding with Google I/O is factually plausible, though the specific claims about developer experiences and AI-assisted code generation require viewing the full article for verification.
Claim Tracker
AI-assessed
Google I/O 2024 confirmed for May 14-16, not May 19. This claim appears inaccurate regarding timing.
Technical feature claim requires verification against Anthropic's official announcements
Broad claim about adoption scale lacks specific data or citations
Specific anecdote sourced from reported event; no direct evidence provided
Ask AI about this story
// discussion
sign in to join the discussion