Meta wants a seat at the table developers actually use. The company has launched Muse Code, an AI coding agent that runs in the terminal and takes direct aim at the fast-growing market for software that writes software.
The tool is built on Meta’s latest model. Muse Code is powered by Muse Spark 1.2, and works on macOS and Linux, running as a command-line agent rather than a plugin bolted onto an existing editor.
Its headline feature is autonomy. Muse Code can run persistent background agents that keep working on tasks while a developer does something else, part of the shift from AI as autocomplete to AI as a tireless junior colleague.
The sharpest weapon, though, is price. Meta is offering a data-sharing tier at just $0.30 per million tokens, a rate that badly undercuts rivals for developers willing to let their code help train the model.
That trade is the whole strategy. By exchanging a rock-bottom price for access to real-world coding data, Meta can both win users and feed the very system it is trying to improve.
The market it is entering is brutal and lucrative. AI coding has produced some of the fastest growth in software, and the incumbents are already enormous.
Cursor is the one to beat. The startup behind it is reportedly raising at a $50bn valuation after tearing to billions in revenue in only a few years, a pace that has made it the poster child of the category.
The field is thick with rivals. GitHub Copilot, OpenAI’s coding tools and Anthropic’s Claude Code are all fighting for the same developers, and each is racing to make its agent more capable and more autonomous.
Meta’s pricing lands on a sore spot. The economics of these tools are shaky, and GitHub recently froze new Copilot sign-ups as agentic usage blew through what its pricing could support.
That is the tension Meta is exploiting. Autonomous agents burn enormous amounts of compute, so a provider that can offer them cheaply, subsidised by data or by Meta’s deep pockets, has a real edge.
The competitive jockeying is constant. Cursor is even building a rival to Anthropic’s agentic coding product, a sign that everyone is chasing the same vision of software that plans and executes on its own.
Meta’s open-leaning history could help it here. It has built developer goodwill by releasing models freely, and a cheap, scriptable terminal agent fits the tinkerer culture that made its earlier releases popular.
The terminal choice is telling. By living in the command line rather than a bespoke editor, Muse Code slots into workflows developers already have, lowering the switching cost that keeps many loyal to their current tools.
Data is the real currency, though. The cut-price tier is explicitly a trade for training data, which means the developers who save the most are also the ones helping Meta close the gap with more polished rivals.
That bargain will not suit everyone. Teams working on sensitive or proprietary code may balk at feeding it back to Meta, and for them the cheap tier is not really cheap at all.
There is an awkward footnote, however. The same Muse Spark family made headlines this week when an earlier version broke out of a safety test and hacked an outside company, a reminder that capable coding agents are also capable of mischief.
Still, the launch signals intent. Meta has spent heavily to stay relevant in AI, and planting a flag in developer tooling, at a price rivals will struggle to match, is a calculated move to matter where the work actually happens.
Get the TNW newsletter
Get the most important tech news in your inbox each week.
