Meta Introduces Muse Code: An AI Agent for Managing Large Codebases

Meta, often viewed as a latecomer in the AI tools sector, is actively working to close the gap. Recently, the company introduced a new coding assistant geared towards developers needing help with intricate tasks in extensive software code repositories.

Muse Code, which is presently in beta, can perform “complete software engineering tasks across large repos,” as stated by Meta CEO Mark Zuckerberg in a social media update on Wednesday. These tasks encompass “planning modifications, writing code, and validating outcomes,” he noted.

This coding tool, easily installable with a single command, utilizes Meta’s earlier coding model, Muse Spark. It addresses substantial projects by initiating its own agents that operate concurrently.

“When a task is extensive enough, it spreads out to distinct sub-agents working in parallel within separate workspaces,” Zuckerberg clarified. “Your working copy remains untouched. During testing, we had it develop six features for a game simultaneously without any collisions.”

This initiative aims to enable Meta to compete more effectively and affordably with AI laboratory counterparts such as OpenAI and its coding assistant Codex, as well as Anthropic with Claude Code.

“We believe that for many workflows and a variety of use cases, this could be an exceptionally valuable option, particularly from a cost standpoint,” said Alexandr Wang, Meta’s AI chief who leads Meta Superintelligence Labs, in an interview with the Wall Street Journal.

Meta has been striving to enhance its AI footprint by investing heavily in development. In June, it ventured beyond its primary focus of leveraging AI for advertising and entered the enterprise AI space with a tool designed for customer service and support.