What is Muse Code, Meta's new AI agent for large codebases
Meta has announced Muse Code, a new AI agent designed to help developers work with large and complex software repositories. The company says the tool aims to address the biggest weakness of earlier-generation coding assistants: their inability to understand the context of an entire codebase.
Most existing AI coding tools can only process a handful of files, or a limited context window, at a time. That means when developers work in large repositories containing thousands of files, suggestion quality tends to drop and the tool can propose inconsistent changes.
Muse Code, according to Meta, works by mapping the dependencies between files across a repository and dynamically updating that map during a query. This approach is designed to let the tool better predict how a change made in one function might ripple elsewhere in the project.
The company claims the tool performs especially well on large-scale refactoring tasks — restructuring code without changing its behavior. These tasks have traditionally been particularly challenging for AI assistants, since they require consistent changes across numerous files simultaneously.
Meta says Muse Code has been tested internally for months and is already used across the company's own large-scale codebases. The company argues this internal use helped the tool mature under real-world conditions before its wider release.
The tool integrates with Meta's previously announced Muse Spark platform, an integration the company says allows code suggestions to draw on its broader AI infrastructure. Meta states the two products are designed to work together to boost developer productivity.
Rival companies are developing similar 'repository-level awareness' approaches; the industry overall is shifting from single-file-focused coding assistants toward tools that understand full project context.
Security experts note that this kind of tool requiring broad access to large codebases can introduce new risks, particularly concerns about sensitive or proprietary code leaking into external systems. Meta emphasizes that data stays within the company and is not transferred externally.
Developer communities have not yet independently tested the tool's real-world performance; early access is currently limited to a small group of external developers. Meta says it plans to gradually expand access based on feedback.
Analysts note that AI tools capable of understanding large codebases could deliver a meaningful productivity boost in software development, though the reliability of these tools at scale still needs to be proven.
Read next
You can now turn off Google Gemini's visible AI watermarks. Here is what that actually changes
Google now lets users toggle off the visible watermark that normally appears on images, videos and music generated with Gemini and its Nano Banana and Omni models. The invisible SynthID watermark and embedded metadata remain in place regardless, meaning the content stays technically traceable as AI-generated even when it no longer looks that way.
A screen-sharing flaw is letting attackers take full control of Macs. Here is what to know
Security researchers have confirmed active exploitation of a vulnerability that lets remote attackers log into Macs without a password through a flaw in the screen-sharing feature. Here is how the exploit works, which systems are exposed, and the steps Mac users should take while a patch is developed.
OpenAI and Anthropic cut prices as Chinese AI rivals gain ground
OpenAI and Anthropic have both released cheaper models in recent weeks, a shift analysts attribute to intensifying competition from Chinese AI labs offering comparable performance at a fraction of the cost. The pricing pressure marks a notable change for two companies that have spent years competing primarily on capability rather than cost.
What we know about the alleged Iranian hacks on US water utilities
Over the past several weeks, hackers have broken into the industrial control systems of multiple US water utilities in an operation researchers attribute to actors linked to the Iranian government. Here is what has been confirmed about the intrusions, the vulnerable equipment involved, and why water infrastructure keeps turning up as a target.
Self-driving trucks are now testing on California highways: what the new permits actually allow
Aurora Innovation and Kodiak AI have received permits from the California Department of Motor Vehicles to test self-driving trucks on the state's highways, a milestone for an industry that has struggled to gain regulatory footing in one of the country's largest freight markets. Here is what the permits do and do not allow.