Meta Releases Muse Code — an AI Agent for Large Repositories

Meta, often regarded as a laggard when it comes to AI tools for developers, is trying to make up lost ground. This week the company released Muse Code — a terminal-based agent aimed at programmers who need help with complex tasks in sizable codebases.

According to Meta chief Mark Zuckerberg, Muse Code is capable of performing “full-fledged engineering tasks in large repositories” — from planning changes and writing code to verifying the result. The tool installs with a single command and runs on the company's previously unveiled Muse Spark model. The product is available in beta for now.

Parallel sub-agents

The main feature Meta highlights is the ability to break a large task into pieces. “When a job is big enough, it gets distributed among separate sub-agents that work in parallel in isolated working copies,” Zuckerberg explained. Meanwhile, the user's main working copy remains untouched. As an example, he cited a test in which the agent simultaneously implemented six features for a game without conflicting with itself.

Betting on price

The launch is Meta's attempt to line up alongside its competitors: OpenAI's Codex agent and Anthropic's Claude Code. The company points to cost as its main selling point. “We think that for many scenarios and workflows this could be an exceptionally good option, especially in terms of cost,” said Alexandr Wang, who heads Meta Superintelligence Labs, in a conversation with the Wall Street Journal.

Meta is expanding its presence in the AI market, pouring substantial funds into development. In June the company moved beyond its core focus — applying AI to its advertising business — and unveiled a corporate agent for customer support, stepping into the enterprise-solutions market.

Source: TechCrunch