Meta Launches Muse Code Terminal Agent and Muse Spark 1.2 to Challenge OpenAI and Anthropic

On August 5–6, Meta re-entered the frontier conversation with a coding push. Muse Code, announced by Mark Zuckerberg ('finkd') and retweeted by both Meta AI and Chief AI Scientist Yann LeCun, is a terminal coding agent that 'takes on complete software engineering tasks across large repos' — a direct shot at Anthropic's Claude Code, OpenAI's coding agents, and the broader terminal-agent category. It shipped in beta alongside a Muse Spark 1.2 model update.
Meta explicitly framed the release as competitive positioning against Anthropic's Opus 5, GPT-5.6 Terra, Gemini 3.6 Flash and Grok 4.5 — an unusually crowded target list that signals Meta is trying to reassert relevance after its open-source Llama models struggled and after tens of billions in GPU capital expenditure. The Muse rebrand (Muse Spark, now Muse Code) reflects Alexandr Wang's Superintelligence Labs reshaping Meta's model identity away from the Llama name.
Separately in the same week, Meta unveiled an agent architecture in which a secondary 'memory coach' AI agent's sole role is managing a primary agent's memory to prevent drift on long-running tasks. That resonated with developers frustrated by agents 'remembering' via stale prompt-stuffed summaries, spawning community projects like Perseus and Perseus Vault. It's also an implicit admission that frontier models can't yet sustain coherent reasoning over multi-day workflows — a candid framing that stood out.
The skeptical context: Meta is spending enormously and shipping fast, but its credibility took a hit with Llama, and a coding agent released into a market already served by mature Claude Code and Cursor integrations has to prove it on real repos, not launch tweets. India's MeitY separately summoned Meta executives for August 5–6 meetings on child safety and AI-content detection, a reminder that Meta's regulatory surface is broad. Watch whether Muse Code's beta metrics — task completion on large repos — hold up against Anthropic's incumbency.