Meta personal AI agents are moving toward the front of Mark Zuckerberg’s product plan. On Meta’s Q2 2026 earnings call Wednesday, the CEO said the company will soon have more to say about agents that can act for users, framing them as a consumer product rather than a tool reserved for coders and office automation pilots.
Zuckerberg described agents that could run around the clock on a person’s behalf to help with goals involving life, health, relationships, finances, and other areas. He said coding has been the first major use case where agents have taken off, partly because engineers are willing to spend time getting brittle tools to behave. For Meta’s version to work at consumer scale, he said, the product needs to work out of the box and be easy enough for billions of people to use.
That is the pitch. The hard part is everything after the pitch.
What are Meta personal AI agents?
An AI agent is software that takes a user’s instruction, breaks it into steps, and tries to complete tasks using an AI model and connected tools. The difference between a chatbot and an agent is action: a chatbot mainly replies, while an agent may attempt to do something for you, depending on what access it has been given. If you want the lower-level machinery, Kernel has an explainer on how LLMs answer prompts.
Meta has not shown the personal agents Zuckerberg discussed on the call, and he did not give a launch date. He said Meta would share more “soon,” and added that personal agents would be “the foundation” for the company’s next set of products and revenue lines over the coming months and years.
The positioning puts Meta near a crowded fight. Anthropic and OpenAI have put significant effort into agents for coding and business work. Google has also pushed personal AI agents through Gemini Spark. The Verge reported that it found Google’s version strong, though unsettling, and said it is not yet clear how Meta’s approach will compare.
Why Meta has a harder sell than Google or Microsoft
The Verge also pointed to several problems Meta will have to handle. Meta’s AI efforts have lagged the larger AI labs, according to that report. Meta also lacks the kind of default access Google and Microsoft have to users’ email, documents, and workplace files, which are exactly the places an assistant would need to read and act if it is supposed to be useful.
Trust is the other obvious obstacle. The Verge noted that Meta is already facing privacy concerns around its smart glasses. Asking people to hand an agent permission to act on their behalf is a larger request than asking them to chat with a bot. Meta can call it personal AI, but users will still ask what data it sees, what it can change, and who gets blamed when it does the wrong thing.
Zuckerberg also used the call to talk up Meta’s existing business agents. He said more than 1 million businesses use the company’s agents on WhatsApp and Messenger each week. Those business agents are also being rolled out to Instagram, he said.
The personal-agent push follows Meta’s Muse Spark AI model release last quarter and a 1.1 update that The Verge reported improved coding capabilities. Meta has reorganized around AI, with The New York Times reporting that about 7,000 employees were reassigned to new AI initiatives. Bloomberg reported separately that Meta laid off about 8,000 people in May.
The spending line is even less subtle. In its Q2 2026 results, Meta said it expects 2026 capital expenditures of $130 billion to $145 billion. The company also announced a 1-gigawatt data center campus with BlackRock the day before the earnings call. Zuckerberg said on the same call that Instagram now has more than 2 billion daily active users.
This story draws on original reporting from The Verge.
