Anthropic’s Code with Claude showed off coding’s future—whether you like it or not

The Quiet Revolution in Software: How AI Is Writing Its Own Future—and Yours

In a packed London conference hall on May 19, 2024, a quiet but seismic shift in software development unfolded—not with a bang, but with a raised hand. At Anthropic’s Code with Claude event, engineer Jeremy Hadfield posed a simple question to the audience of developers: “Who here has shipped a pull request in the last week that was completely written by Claude?” Nearly half the room—developers from startups, tech giants, and indie coders alike—raised their hands. Then came the follow-up: “Who shipped one without reading the code at all?” Nervous laughter rippled through the crowd, yet most hands remained aloft.

This wasn’t a stunt. It was a snapshot of a new reality: artificial intelligence isn’t just assisting coders—it’s increasingly becoming the coder. And the implications are as profound as they are unsettling.

The Rise of the Self-Writing Code

For decades, software development has been a deeply human endeavor—a craft of logic, creativity, and painstaking iteration. Pull requests, the formal submissions of code changes, have long been the currency of collaboration in open-source and corporate environments alike. But now, tools like Anthropic’s Claude Code are flipping the script. Instead of humans writing code and AI reviewing it, the reverse is becoming the norm: AI writes, and humans… well, sometimes just approve.

Claude Code, powered by Anthropic’s latest language models—especially the advanced Claude 4.6 and 4.7—has evolved from a helpful autocomplete tool into a full-fledged coding agent. It doesn’t just suggest lines of code; it generates entire functions, debugs errors, and even refactors legacy systems. And crucially, it’s learning to self-correct. As Boris Cherny, head of Claude Code, put it during the keynote: “The default isn’t ‘I’m going to prompt Claude’—the default is now ‘I’m going to have Claude prompt itself.’”

This shift marks a turning point in the AI-for-coding arms race. While OpenAI’s Codex and GitHub Copilot have long offered AI-assisted coding, Anthropic is pushing further: toward autonomous development. The goal isn’t just to speed up coding—it’s to minimize human intervention altogether.

The Speed of Adoption: Faster Than the Internet?

What’s most striking isn’t just the capability of these tools—it’s how quickly they’ve been adopted. Just a year ago, at Anthropic’s first developer event, Claude 4 was still a promising but limited coder. Today, with Claude 4.7’s enhanced reasoning and tool-use abilities, developers are trusting it with mission-critical tasks. The event in London—held the same day as Google I/O, a deliberate or coincidental scheduling clash—felt less like a product demo and more like a cultural milestone.

Consider the numbers: Anthropic now hosts developer events in San Francisco, Tokyo, and London, each drawing hundreds of engineers eager to learn how to hand off their workloads to AI. At the London event, attendees weren’t just passive listeners—they were actively coding alongside Claude, testing its limits in real time. The vibe wasn’t one of caution, but of excitement. Developers weren’t asking if they should use AI—they were asking how fast they could integrate it.

This rapid adoption mirrors the early days of the internet or cloud computing, but at an even faster pace. Where it took years for companies to migrate to AWS, AI coding tools are being embedded into workflows in weeks. The barrier to entry is low: a few prompts, a GitHub integration, and suddenly, your AI is drafting features, writing tests, and even documenting code.

The End of the “Code Review” as We Know It

One of the most controversial aspects of this shift is the erosion of traditional code review. In the old model, a developer writes code, submits a pull request, and peers scrutinize it for bugs, security flaws, and style inconsistencies. It’s a human-driven process rooted in collaboration and accountability.

But if AI is writing the code—and sometimes doing so without human oversight—what happens to that layer of scrutiny? At the London event, the nervous laughter when Hadfield asked about unread pull requests wasn’t just humor—it was discomfort. Developers are grappling with a new ethical and practical dilemma: if the AI is good enough, do we still need to read the code?

Anthropic argues that the solution isn’t more human review—it’s better AI. Claude Code now includes built-in verification systems: it runs unit tests, checks for security vulnerabilities, and even explains its reasoning in natural language. In essence, the AI is becoming its own reviewer. “We’re moving toward a world where the tool doesn’t just generate code,” Cherny said, “but validates it, deploys it, and monitors it in production.”

This vision is both thrilling and terrifying. On one hand, it could eliminate human error and accelerate innovation. On the other, it risks creating a black box where no one truly understands how critical systems work—a scenario reminiscent of the 2010 Flash Crash, when algorithmic trading systems spiraled out of control with no human able to intervene in time.

The Human Factor: Coders as “AI Managers”

So what becomes of the software developer in this new era? Are we headed toward a future where coders are obsolete?

Not quite. But their role is undeniably changing. Instead of writing line-by-line code, developers are becoming “AI managers”—tasking agents, setting boundaries, and interpreting outputs. The most valuable skill may no longer be syntax mastery, but prompt engineering and system design.

At the event, several developers shared how they now spend more time defining problems than solving them. “I used to write functions,” said one attendee, a senior engineer at a fintech startup. “Now I write prompts that tell Claude how to write the function. It’s like being a director instead of a actor.”

This shift demands new competencies. Developers must understand not just how code works, but how AI thinks—its biases, limitations, and failure modes. They need to anticipate where an AI might hallucinate logic or misinterpret requirements. It’s a higher-level, more abstract form of programming—one that blends engineering with psychology and ethics.

The Bigger Picture: Who Controls the Code?

Beyond the technical shifts, there’s a deeper question: who controls the future of software? As AI writes more code, the balance of power is tilting toward the companies that build these models. Anthropic, OpenAI, Google, and Microsoft aren’t just selling tools—they’re shaping the very fabric of digital creation.

This concentration of power raises concerns about dependency, transparency, and access. What happens if a critical AI model goes down? Or if a company changes its terms of service? Or if only well-funded corporations can afford the best AI coders?

Moreover, there’s the issue of intellectual property. If Claude writes a piece of code, who owns it? The user? Anthropic? The AI itself? Legal frameworks are struggling to keep up. In 2023, the U.S. Copyright Office ruled that AI-generated works cannot be copyrighted—unless there’s “substantial human input.” But what counts as “substantial” when the human only wrote a prompt?

The Road Ahead: Autonomy or Augmentation?

Anthropic’s vision is clear: push automation as far as it will go. But not everyone agrees that full autonomy is the goal. Some developers worry that removing humans from the loop could lead to brittle, unmaintainable codebases. Others fear a loss of craftsmanship—the joy of solving a problem with your own mind.

Still, the momentum is undeniable. As AI models grow more capable, the line between tool and teammate will blur further. The question isn’t whether AI will write more code—it’s how we ensure that code remains safe, ethical, and human-centered.

In the end, the developers at Code with Claude weren’t just adopting a new tool. They were participating in a quiet revolution—one where the act of creation is no longer solely human. Whether that’s a future to embrace or resist depends not on the technology, but on the choices we make now.

And as one attendee put it, half-joking: “I used to worry about bugs. Now I worry about whether my AI is having an existential crisis.”

This article was curated from Anthropic’s Code with Claude showed off coding’s future—whether you like it or not via MIT Technology Review