Can Meta close the gap with OpenAI and Anthropic in coding agents?

Meta enters a crowded field

Meta has introduced Muse Code, a beta terminal coding agent for programmers working in large software code bases, according to TechCrunch. Mark Zuckerberg said on social media that the tool can handle complete software engineering tasks across large repos, including planning changes, writing code, and validating results.

That puts Meta in direct competition with OpenAI’s Codex and Anthropic’s Claude Code, two products TechCrunch said Meta wants Muse Code to challenge. Alexandr Wang, who leads Meta Superintelligence Labs, told the Wall Street Journal that Meta sees the tool as a lower-cost option for some workflows and use cases, according to TechCrunch’s report.

The pitch is straightforward. Meta is not trying to sell a general chatbot with coding features bolted on. It is trying to sell a coding agent built for large repositories, with a workflow centered on parallel sub-agents and isolated worktrees.

What Muse Code does

TechCrunch reported that Muse Code is powered by Meta’s Muse Spark model and can spin up sub-agents that work in parallel while leaving the user’s working copy untouched. Zuckerberg said testing found the system could build six game features at once without collisions.

That architecture matters because large code bases punish sloppy automation. A coding tool that can split work into isolated branches may appeal to teams that already spend time managing merge conflicts, review cycles, and partial failures. Meta is using that pain point as the entry point.

The company has not, at least in the reporting available here, published independent benchmark results for Muse Code. TechCrunch also did not report pricing, licensing terms, or customer adoption data. That leaves Meta asking buyers to take the company’s claims on faith for now.

For enterprise buyers, that is a thin basis for procurement decisions. For product leaders, it means the public case for Muse Code rests on product direction rather than measured performance.

The enterprise push behind it

Muse Code is not arriving in isolation. TechCrunch reported that Meta entered the enterprise AI market in June with an agent aimed at customer service and support. That move widened the company’s push beyond ad-support tools and into business software.

The timing matters. Meta has been spending heavily on AI development, according to TechCrunch, and the company now appears to be testing whether that spending can produce a second line of business outside advertising. Muse Code fits that effort by giving Meta another sales story for corporate buyers.

The company’s challenge is less about building one more product than about proving it can sell into workflows where OpenAI and Anthropic already carry stronger brand recognition. Coding agents are not a side category. For many software teams, they are becoming a daily tool, and the vendors that get in early can shape habits that are hard to dislodge later.

Meta’s late arrival

The question is not whether Meta can build a coding agent. It clearly can. The question is whether Muse Code arrived early enough to matter.

TechCrunch’s report suggests the market is already moving. In its July coverage of Nous Research, TechCrunch said the maker of the open-source Hermes agent was in talks for at least $75 million at a $1.5 billion valuation, with investor interest from Robot Ventures, USV, and others. The same report said Hermes had roughly 214,000 GitHub stars and nearly 40,000 forks, and that it was available locally and as a hosted service with paid tiers from $20 to $200 a month.

That is not the profile of a sleepy category. It points to a field that is already crowded with open-source projects, startup-backed tools, and large commercial platforms. Meta is entering after the first wave of attention, not before it.

OpenAI and Anthropic also have something Meta does not yet have in this category: a default place in the minds of many developers looking for coding help. The product names are already familiar in software teams. Meta has to spend time convincing buyers that it belongs in the same conversation.

Cost may be Meta’s opening

If Meta has a path into this market, cost may be it. Wang’s comment to the Wall Street Journal, as reported by TechCrunch, framed Muse Code as a lower-cost option for some workflows and use cases.

That matters because enterprise software buyers do not need every tool to win on raw capability. They need tools that fit a budget and slot into existing processes. If Meta can price Muse Code below rivals while handling enough of the work well enough, it could win a narrow but real foothold.

Still, price alone does not settle the case. Buyers will ask how Muse Code performs on large repositories, how much supervision it needs, and whether it can handle real production work without creating cleanup work for engineers. TechCrunch did not report answers to those questions.

Until Meta releases more evidence, Muse Code looks like a strategic bet rather than a proven product. It gives Meta another enterprise story, another AI surface area, and another chance to turn spending into revenue. It does not yet give the company proof that it can pull developers away from OpenAI or Anthropic.

What to watch next

The next tests are practical, not rhetorical. Meta has not said when Muse Code will leave beta, TechCrunch reported, and the company has not disclosed which customers have access now. The reporting also does not say whether Muse Code will be tied to other developer products already inside Meta’s software stack.

Those details will decide whether Muse Code becomes a real challenger or just another name in a fast-moving market. For now, Meta has entered the race. It has not shown that it can close the distance.