Meta Launches Muse Code, a Terminal-Based AI Agent for Large Codebases

Meta has released Muse Code, a terminal-based AI coding agent designed to handle software engineering tasks across large code repositories. The tool is currently available in beta and can be installed with a single command.

CEO Mark Zuckerberg announced the launch in a social media post on Wednesday, saying Muse Code can "complete software engineering tasks across large repos," including "planning changes, writing code, validating the results."

How it works

Muse Code is powered by Muse Spark, Meta's previously released coding model. The agent's key architectural feature is its ability to parallelize work on large jobs. According to Zuckerberg, "When a job is big enough, it fans out to separate sub-agents working in parallel in isolated worktrees. Your working copy is never touched." Meta says internal testing had the system build six features for a game simultaneously without collisions between sub-agents.

No pricing has been disclosed yet, though Meta AI chief Alexandr Wang, who leads Meta Superintelligence Labs, told the Wall Street Journal that cost will be a central selling point: "We think that for a lot of workflows and a lot of use cases, this can be an incredibly good option, especially from a cost perspective."

Competitive context

The launch is Meta's attempt to gain ground in the AI coding-agent market, where it has trailed competitors. OpenAI's Codex and Anthropic's Claude Code have established significant developer mindshare in the terminal-agent category, alongside tools from companies like Anysphere, Sourcegraph, and GitHub. Meta's entry adds a fourth major lab-backed option to a field that has grown crowded over the past two years.

This release also extends Meta's broader push to diversify its AI business beyond advertising. In June, the company entered the enterprise AI market with an agent aimed at customer service and support. Muse Code represents a parallel move into developer tools, an area where Meta has historically had less commercial presence compared to its research-focused releases like the Llama model family.

No specific benchmark scores, context window size, or detailed pricing tiers were disclosed in Meta's announcement.

What this means

Meta's entry into terminal coding agents is notable less for technical novelty — parallel sub-agent orchestration has precedent in tools from Anthropic and others — and more for competitive positioning. By emphasizing cost over raw capability, Meta appears to be targeting price-sensitive developers and teams rather than competing head-on with Codex or Claude Code on top-line benchmark performance.

The lack of disclosed pricing, context window specs, or benchmark comparisons makes it difficult to assess how Muse Code stacks up technically against incumbents. Given Meta's history of using aggressive pricing and open distribution to gain developer adoption (as with Llama), a cost-first strategy for Muse Code would be consistent with the company's broader playbook. Whether that's enough to win developer mindshare in a market where Anthropic and OpenAI already have entrenched workflows remains an open question.