Editorial illustration for Meta Adds Persistent Async Agents to Muse AI Coding Tools
Meta Launches Async Agents for Muse AI Coding
Meta released two AI coding tools on Wednesday: Muse Code, a terminal-based coding agent now in beta, and Muse Spark 1.2, an update to its Muse Spark model family tuned for software work. The pairing puts Meta squarely against Anthropic's Claude Code and OpenAI's Codex, tools that have already reshaped how professional developers write and ship software. Startups like Cursor built billion-dollar valuations on this same workflow, and Meta had mostly stayed out of it, letting its Llama models rack up more than a billion downloads as an open-weight alternative instead of a packaged coding product.
Muse Code breaks from that pattern. It's a full harness that installs on macOS or Linux with one curl command, built to run alongside the Muse Spark model that powers it. Unlike Llama, both pieces are proprietary.
Installation requires a Meta account and login credentials before the agent can start working across a codebase. Mark Zuckerberg announced the beta himself, posting from his X account under the handle @finkd, framing it as a tool meant to handle full engineering tasks rather than just autocomplete suggestions.
Meta today released Muse Code, a terminal-based AI coding agent now in beta, alongside Muse Spark 1.2, a coding-focused update to its Muse Spark family of frontier models — a one-two punch that puts the company in direct competition with Anthropic's Claude Code, OpenAI's Codex, and the growing field of agentic coding harnesses that have rapidly become the primary way many professional developers ship software.
Why this matters
Meta showing up to the agentic coding fight with a terminal tool and a co-trained model tells us the category has moved past experimentation and into standard infrastructure. Claude Code and Codex already proved developers will hand real repos to an agent that plans, tests, and revises on its own; Muse Code's /plan, /grill, and /goal commands read like Meta studying that playbook closely rather than reinventing it. That's not a knock. Fast-following a proven workflow, backed by a model built specifically for its own harness, is a legitimate strategy when the market's still young.
For developers, the near-term upside is competition pushing down prices and pushing up context handling across all three players. For founders building on top of these tools, the real question is lock-in: whether a harness co-trained with its model traps you the way certain IDEs and clouds already do. We'd watch how Muse Code's approval-gated planning holds up on messy, real-world codebases, not the clean demo repos vendors love to show off. That's where these agents actually earn trust or lose it.
Related Reading
Browse all Meta news →
Business & Startups Nordic pilot adds Gemini for Education, NotebookLM to boost AI literacy
Business & Startups Kling launches Video O1, all-in-one model with MVL bridge using transformer
Business & Startups DeepSeek Seeks More Capital Weeks After USD 7B Funding Round
Related Topic Meta AI Update Pulls From Your Calendar for Daily Briefings
LLMs & Generative AI
Related Topic WhatsApp launches Meta AI Incognito Chat, cuts latency for privacy
AI Tools & Apps
From Archives Google AI leaders leave to launch discovery-focused startup
From Archives Kavukcuoglu Takes Over Google DeepMind After Leadership Exits
From Archives Meta AI Launches Muse Code Beta, Powered by Muse Spark 1.2
Open Source
From Archives Meta AI’s Memory Coach Outperforms Constant Recall for Long Tasks
Research & Benchmarks
Common Questions Answered
What are the two new AI coding tools that Meta released?
Meta released Muse Code, a terminal-based coding agent now in beta, and Muse Spark 1.2, an update to its Muse Spark model family specifically tuned for software work. These tools are designed to compete directly with existing solutions like Anthropic's Claude Code and OpenAI's Codex in the agentic coding space.
How does Muse Code compare to competitors like Claude Code and Codex?
Muse Code features terminal-based commands like /plan, /grill, and /goal that demonstrate Meta studied the proven workflow established by Claude Code and Codex, which allow developers to hand real repositories to agents that plan, test, and revise code autonomously. Rather than reinventing the category, Meta is fast-following a proven approach backed by its co-trained Muse Spark 1.2 model.
Why is Meta's entry into agentic coding tools significant?
Meta's release of persistent async agents through Muse Code and Muse Spark 1.2 signals that agentic coding has moved beyond experimentation into standard infrastructure for professional developers. The category has already proven its value through competitors and startups like Cursor that built billion-dollar valuations on this workflow, making Meta's direct competition in this space a major strategic move.
What commands does Muse Code provide for developers?
Muse Code includes terminal-based commands such as /plan, /grill, and /goal that enable developers to interact with the AI coding agent. These commands allow the agent to plan code changes, test implementations, and work toward specific development goals autonomously.
Further Reading
- Meta enters the AI coding wars with Muse Spark 1.2 and Muse Code with persistent async background agents - VentureBeat
- Meta launches Muse Code AI coding agent for macOS and Linux - 9to5Mac
- Meta debuts first AI coding agent to take on Anthropic and OpenAI - CNBC
- Meta jumps into AI coding market to chase Anthropic and OpenAI - CNBC
- Introducing Muse Spark 1.1 - Meta AI Blog
