Meta ships Muse Code, a terminal agent to fight Claude Code

Meta launched Muse Code, a terminal-native coding agent running on its new Muse Spark 1.2 model — a direct swing at Anthropic's Claude Code and OpenAI's Codex. Meta has been conspicuously absent from the dev-tools fight while rivals turned coding agents into their stickiest surface; shipping into the terminal is the cheapest way to buy back developer mindshare.

Liquid AI went the other direction: LFM2.5-2.6B is a 2.6B-parameter on-device agent that plans and calls tools at roughly 220 tokens/sec on phones and laptops, no cloud round-trip. Small models that can actually use tools are the thing that makes local agents useful rather than a demo.

And Anthropic hired former California Supreme Court Justice Tino Cuéllar as its first Chief Global Affairs Officer — a judge, not a lobbyist, running global AI policy. That's a bet that the next two years of AI regulation get argued in legal terms, not just political ones.

Why it matters: the agent race is splitting into three fronts — the terminal, the device, and the statehouse — and the big labs are now staffing all three.