The AI coding agent track just got another heavyweight entrant.

On August 5, Meta officially launched Muse Code — a terminal AI coding agent currently in public beta. Installable with a single command, it handles complete software engineering tasks across large code repositories: planning changes, writing code, validating results — the entire workflow automated.

The product's technical signature is interesting: multi-agent parallelism. When a task is big enough, Muse Code automatically splits it into subtasks and spins up multiple sub-agents working in parallel in isolated worktrees. They don't interfere with each other, and your working copy is never touched. Zuckerberg personally endorsed it: in testing, we had it build six features for a game simultaneously with zero collisions.

Powering Muse Code is Muse Spark, Meta's previously released coding model. And leading the product is Alexandr Wang — the former CEO of Scale AI, who joined Meta earlier this year and now heads Meta Superintelligence Labs.

The Price Slayer Enters

Meta's approach to entering the AI coding race is the same as its approach to every other AI race: lead with price.

Currently the top players in AI coding agents are OpenAI Codex and Anthropic Claude Code. Take Claude Code Sonnet, for example: input pricing is roughly $3 per million tokens. Muse Code's input pricing is $1.25 per million tokens — roughly 42% of Claude Code Sonnet's price.

Near half the price, without a significant drop in capability. Alexandr Wang put it bluntly in a Wall Street Journal interview: "We think that for a lot of workflows and a lot of use cases, this can be an incredibly good option, especially from a cost perspective."

Translation: we're cheap, and we're good enough. For many small teams and individual developers, good enough + cheap is all that matters.

Meta already validated this playbook with open-source LLMs. The Llama series, backed by open source + free + performance approaching closed-source models, quickly took the top spot in the open-source LLM ecosystem. Now it wants to use the same strategy on the AI coding market: performance isn't that far behind, but the price is half — how do you compete with that?

The AI Coding Price War Has Officially Begun

The release of Muse Code marks one thing: the AI coding tool price war has officially started.

Over the past year, AI coding assistants have evolved from the Copilot generation-suggestion model to full agent-based workflows. Codex, Claude Code, Cursor, Devin, Windsurf… more and more products, stronger and stronger capabilities, but prices have also gone up. Top-tier coding agents cost tens or hundreds of dollars a month, which isn't cheap for individual developers and small teams.

With Meta entering the picture, this dynamic gets disrupted. It has compute, models, and an open-source ecosystem — nobody can beat it on price. More importantly, Meta isn't building AI coding agents to make money from them — it's building ecosystem, attracting developers, and completing its AI toolchain. As long as it keeps developers in Meta's AI ecosystem, it doesn't matter if Muse Code doesn't turn a profit on its own.

For developers, this is obviously great news. More choices, lower prices, better tools. But for other AI coding startups, the pressure just got real — when Meta drives prices down to this level, you either cut prices (thinner margins), differentiate (very hard), or die.

AI coding is one of the fastest-adopting and highest-willingness-to-pay sectors in AI. But with giants entering the ring, the startup window for this track is closing fast. The AI coding companies that survive will either have exceptional product experience and user stickiness, or irreplaceable vertical capabilities — otherwise, facing the giants' price offensive, it's really hard to hold on.

See you tomorrow.