The first full week of August brought a new heavyweight into coding agents, the year's biggest lab-leadership shakeup, and a burst of funding for agents that do — and secure — real work.
Meta ships a terminal coding agent
On August 5, Meta released Muse Code in beta for macOS and Linux — a terminal-based coding agent powered by its Muse Spark 1.2 model, aimed squarely at Anthropic's Claude Code and OpenAI's Codex. It plans changes, writes and validates code, and spawns parallel sub-agents in isolated worktrees; Meta demoed it building six game features simultaneously without collisions, and AI chief Alexandr Wang pitched it as "an incredibly good option, especially from a cost perspective." A price-aggressive third giant entering the category pressures everyone's margins — see how the field stacks up on the agentic coding board.
DeepMind reorganizes; its veterans go agentic
Also on August 5, Demis Hassabis stepped back from day-to-day DeepMind leadership to become Chairman of Google DeepMind and Alphabet Chief Scientist, with CTO Koray Kavukcuoglu taking over daily operations. The same day, 27-year Google veteran Jeff Dean departed with Sanjay Ghemawat, Oriol Vinyals, and Quoc Le to co-found Discovery Loop, a public-benefit corporation automating ML, science, and engineering — with Alphabet itself among the seed backers alongside Radical, Khosla, Lightspeed, and Kleiner Perkins. The people who built modern deep learning are now betting on agents that automate research itself.
A ~$360M week for agents in production — and agent security
Three raises told one story. HappyRobot, whose voice and text agents automate freight-brokerage calls and paperwork, closed a $150M Series C at a $1.22B valuation (August 4) — 8 of the 10 largest US freight brokers run its agents in production. On the security side, Zenity raised a $125M Series C led by Norwest (August 3) to govern enterprise AI agents, and Obsidian Security added $85M at a $1.1B valuation (August 4) for governing agents across third-party SaaS. Boring vertical voice work is minting unicorns, and securing what agents do is now its own category — compare voice AI and automation & agents.
Model watch: unlimited GPT-5.6 for free users
In ChatGPT updates rolling out the week of August 3–6, OpenAI made GPT-5.6 Luna the default for Free and Go users with unlimited text chats and a new "Think" button, and upgraded GPT-5.6 Sol for paid tiers with better factual reliability plus a reasoning-effort slider — OpenAI says responses with at least one factual error were ~62% less common with Luna than its predecessor on finance, medical, and legal prompts. Unlimited frontier-tier chat at $0 resets the baseline every consumer AI app competes against.
What it means if you're picking tools
The agentic coding category just gained a third giant, and the smart money is flowing to agents doing unglamorous production work — plus the tooling to secure them. If you're evaluating automation platforms or voice agents, the field is deepening fast. Vote for what's actually delivering.
Sources: Meta Muse Code (TechCrunch) · DeepMind leadership change (Axios) · Hassabis & Dean transitions (The Decoder) · HappyRobot $150M (Businesswire) · Zenity $125M (SecurityWeek) · Obsidian Security $85M (SecurityWeek) · GPT-5.6 Luna free tier (MacRumors) · GPT-5.6 Sol update (OpenAI)