Updated
Updated · MIT Technology Review · May 21
Anthropic Showcases Claude Code 4.7 and Dreaming as Developers Hand Off Pull Requests
Updated
Updated · MIT Technology Review · May 21

Anthropic Showcases Claude Code 4.7 and Dreaming as Developers Hand Off Pull Requests

1 articles · Updated · MIT Technology Review · May 21
  • Nearly half the developers at Anthropic’s London event said they had shipped a pull request entirely written by Claude, and many kept their hands up when asked if they had not read the code.
  • Claude Code’s newer 4.6 and 4.7 releases, rolled out in February and April, are central to Anthropic’s push to let the model test, revise and prompt itself rather than rely on humans to clean up output.
  • The new “dreaming” feature consolidates notes left by Claude coding agents across tasks, aiming to help later agents learn a code base faster and improve over time.
  • Anthropic says old software best practices still apply, but its own managers are struggling to keep up with the surge in AI-generated code as critics warn of weaker oversight, security flaws and eroding developer skills.
  • Executives described Claude as roughly a midlevel engineer today and said the long-term goal is broader automation, with Claude eventually able to build itself.
Is AI-driven coding creating a productivity boom or a hidden technical debt crisis for the future?
Can an AI that 'dreams' and learns on its own be trusted to build our critical software?
As AI writes code developers don't review, who is liable for the resulting security breaches?

Claude Opus 4.7 Sets New AI Coding Benchmark: Autonomy, Code Review, and Industry Impact in 2026

Overview

Anthropic launched Claude Opus 4.7 in April 2026, positioning it as their most powerful model yet. Designed to handle highly complex tasks, Opus 4.7 excels in advanced software engineering and demanding cognitive domains. Its enhanced ability to manage intricate, long-running tasks gives users confidence to delegate challenging coding work that once needed close supervision. The model’s significant gains in reasoning, memory, and visual understanding make it a valuable tool for developers. Alongside the release, Anthropic updated Claude Code and API parameters, streamlining integration and reinforcing Opus 4.7 as the new high-end standard for diverse applications.

...