Meta to Offer Free Personal AI Agents to Billions

Mark Zuckerberg wrote on Aug. 10 that Meta will provide free personal AI agents to billions and is planning up to $145 billion in capital spending in 2026 to support the effort. In the essay he wrote, “Everyone will soon have invention superpowers,” and he added that “Everyone will have an exceptionally capable personal agent that understands you, your goals, and everything you care about.” The agents are described as able to write code, help start and scale businesses and manage household tasks. Meta’s models are built to plan work, connect with applications and carry out multistep actions on a user’s behalf.

Zuckerberg cited examples at the company and at home. He pointed to researchers using AI to generate crystal structures for augmented-reality hardware and to his 8-year-old daughter using tools to turn ideas into videos. The essay says the goal is to lower the time, resources and specialist skills needed for coding, video production, product design and scientific research.

Open-source distribution is central to Meta’s plan. The company released Muse Glimmer, a 30-billion-parameter agentic model under the Apache 2.0 license that it says can run on a Mac or a PC with a single consumer GPU. Muse Glimmer follows earlier models such as Muse Spark, and Meta plans to publish Muse Spark 1.2 weights in the near term. The company argues wider availability of models will reduce concentration of advanced AI capability in a few labs.

On policy, Zuckerberg urged U.S. regulators to ease rules on training data and model distillation for open models and proposed that frontier labs share intermediate training checkpoints with the government. That proposal goes beyond a June executive order that outlined a voluntary early-access framework of up to 30 days.

Meta described governance and privacy steps tied to releases. An independent board will set safety criteria and review compliance so deployment decisions are not concentrated in a single executive. The company outlined a fully private mode intended to prevent Meta or other providers from accessing data handled by a user’s agent. For high-compute use, Meta described dynamic auction pricing as a way to allocate capacity.

Much of the planned capital spending will go to AI infrastructure and data centers. Zuckerberg outlined investments in jobs, schools, energy and public services in communities that host data centers and pledged that “Everyone will have free or affordable access to these tools.” He said broader access could let individuals start and scale businesses with far less capital and staff.

Meta described agent applications beyond household tasks, including agents that interact with blockchains, manage digital assets and execute programmed transactions. The company noted its acquisition of Moltbook, an experimental social network for AI agents with nearly 200,000 verified agents tied to human owners. Meta also said rapid model development is shifting labs from annual flagship launches to updates every few weeks.