Meta launches Muse Code, an AI agent for large code bases
TechCrunch Lucas Ropek ● Covered by 3 sources
Meta just launched Muse Code, a terminal AI agent that can handle coding tasks across huge codebases. It's Meta's bid to compete with OpenAI's Codex and Anthropic's Claude Code, mainly on price.
Meta wants in on the AI coding wars, and it just dropped its entry: Muse Code, a terminal-based agent built to tackle software engineering work across sprawling repositories. Mark Zuckerberg announced the beta release Wednesday, describing an agent that can plan changes, write code, and validate the results, all from a single command-line install.
What sets Muse Code apart, at least on paper, is how it scales. Powered by Meta's existing Muse Spark model, the agent doesn't just chug through one task at a time. When a job is big enough, Zuckerberg said, it splits into sub-agents that run in parallel inside isolated worktrees, leaving the main working copy untouched. He claimed a test run had the system build six separate game features simultaneously without a single collision. That's a real technical flex if it holds up outside of Meta's own demos.
The timing says a lot. Meta has trailed OpenAI's Codex and Anthropic's Claude Code in the coding-agent race, and this launch is clearly aimed at closing that gap. Alexandr Wang, who now runs Meta Superintelligence Labs, told the Wall Street Journal the pitch isn't necessarily raw capability but cost. He framed Muse Code as a strong option for teams that want solid performance without paying premium rates for it.
This fits a broader pattern at Meta, which has spent heavily trying to stretch its AI ambitions beyond advertising. Back in June, the company pushed into enterprise AI with a customer service agent, its first real move outside the ad-tech comfort zone. Muse Code is another step in that expansion, betting that undercutting rivals on price can buy Meta relevance in a market it entered late.
My take
Meta showing up late to the coding-agent party with a cheaper option isn't surprising, it's basically their playbook: let someone else prove the market exists, then compete on price and scale. The parallel sub-agent trick sounds genuinely useful, but until independent developers stress-test it on messy, real-world repos rather than Meta's tidy game demo, the
Read more about this at: TechCrunch
