On May 20, at a venue on the banks of the River Thames in London, Anthropic held its first European developer conference, Code with Claude London.
The host asked the audience: "Who here has submitted a pull request written entirely by Claude this week?" Nearly half the room raised their hands. Then: "And who merged it without reading the code?" Most hands stayed up.
This isn't a joke among programmers
MIT Technology Review covered the scene in a May 21 article bluntly titled "Anthropic showed you the future of coding—whether you like it or not." The attendees included enterprise clients, startup employees, and heavy Claude users—essentially the people closest to the front lines of this technology shift. They are both the customers Anthropic wants to win and the ones directly bearing the impact of this change.
The show of hands sent a stronger signal than any Claude benchmark slide: generating code, skipping review, and merging directly is already a daily reality for many professional developers.
Anthropic's speakers are accelerating this trend
Claude Code lead Boris Cherny stated: "The default is not 'I'll prompt Claude' but 'let Claude prompt itself.'" Researcher Katelyn Lesse added: "I think Claude is roughly at mid-level engineer level right now." Product manager Angela Jiang said: "Our ultimate goal is for Claude to essentially build itself."
That same day, Anthropic launched a feature called "Dreaming," where Claude Code agents write notes for themselves during tasks, so the next agent working on the same project can use them to recognize error patterns and reuse experience. The entire mechanism points in one direction: humans out of the loop.
But opposition exists outside the room
MIT Tech Review directly quoted a Hacker News comment: "The only people who think generating code is no problem are those who never read the code." Other concerns from the community include: long-term erosion of engineers' understanding of their own codebase when they don't read AI-written code; difficulty debugging when you don't know the AI's original reasoning; mass injection of security vulnerabilities because the AI doesn't know which parts of your company are landmines; and a broken pipeline for training senior engineers as junior roles are eaten by Claude, leaving no one to grow into the next Boris Cherny.
Cherny himself added on stage: "There's always room for hand-crafted code, like vegetables at a farmers' market." The analogy is subtle: farmers' market vegetables are a premium choice, but industrial greenhouses are the mainstream—he didn't deny that the mainstream is the latter.
This room represents not the future, but the present
An underlying subtext of the entire event was less optimistic: every time Anthropic releases a new version, it expands the boundary of letting Claude write, review, and revise its own code. Customers are happy to cooperate—half the room raising hands to merge without reading is the strongest evidence of that alignment.
As for whether merging PRs without reading code will cause major problems five years from now, the discussion in the community isn't about "if" but "which company will be hit first." The next time an incident like the 27-year-old OpenBSD bug occurs, we'll have to ask whether it was discovered by AI or originally written in by AI.
Sources: Anthropic's Code with Claude showed off coding's future—whether you like it or not (MIT Technology Review); CocoLoop, Anthropic lands in London as AI-powered coding—and the anxieties around it—go mainstream (Fortune)