Some may have scratched their heads Tuesday morning when news broke that Meta had acquired Maltbook, a social network for AI agents. What on earth does Meta, an ad-supported company, want from a social network whose users are bots? After all, bots are not the target users of brand marketers and advertisers.
Meta doesn’t say much. The only official comment was a short statement that the Moltbook team will join the Meta Superintelligence Lab, opening up “new ways for AI agents to work with people and businesses.”
Reading between the lines, this was an acquisition. A network built for bots is not a natural home for brand advertising, even if Moltbook isn’t fully human. What Meta really wanted was the people behind it, people who enjoyed brainstorming and experimenting with the AI agent ecosystem. And, counterintuitively, that could be a boon for the company’s advertising business.
As Meta CEO Mark Zuckerberg said last year, he believes in a future where “all businesses will soon have business AI in the same way they have email addresses, social media accounts, and websites.” In the agent web, where AI systems work independently on your behalf, AI agents can interact with each other to buy ads, make reservations, serve customers, and more.
AI is also being used to generate ad creative and adjust output based on who is viewing it. AI systems can also manage product pricing and generate personalized offers.
On the consumer side, agents can be used to find the best prices and deals, manage reservations, and purchase products. In some limited cases, agents can already check out and make payments on behalf of consumers. (Agent commerce is still in its infancy, and these systems don’t always work as advertised. But the market moves quickly, and improvements could come quickly.)
Just as Facebook once built a “friend graph” (a network defined by social connections between people, where every individual is a node), the agent web could benefit from an “agent graph”, a system that plans how different agents are connected and what actions they can perform on each other’s behalf.
However, for an agent web that allows enterprise and consumer agents to work together, agents must first be able to find each other, connect, and coordinate their activities. Just as Facebook once built a “friend graph” (a network defined by social connections between people, where every individual is a node), the agent web could benefit from an “agent graph”, a system that plans how different agents are connected and what actions they can perform on each other’s behalf. This can span areas such as travel, online shopping, media and research, and productivity tools.
This could also be the role of ad slots. Today, humans see ads and click on them when they see something of interest, but in an agent web where agents do the shopping for you, ads can look very different. Instead of convincing humans to buy a product, a company’s agents may need to negotiate directly with consumer agents to make the sale.
Perhaps the consumer wants to buy that shirt or lipstick in a certain color and at a certain price. Perhaps the system becomes so complex that these considerations go beyond product and price. Perhaps consumers prefer to support small businesses and stores that only carry environmentally friendly companies. Perhaps consumers only buy a product when it’s on sale, or perhaps they buy a generic version if the ingredients are the same. and so on.
In that case, you will not only need to connect an AI agent, but also rank products by those that best suit individual customer needs. If Meta can tap into that market, AI at the orchestration layer, where the system decides which agents interact with each other and in what order, it could potentially expand its advertising business into entirely new areas.
This all depends on whether consumers actually accept the agent web or trust the AI enough to act on their behalf. But the very existence of OpenClaw, the personal AI assistant that added content to Moltbook, suggests that at least some people are already leaning toward autonomous AI agents.
Of course, there are other possible reasons why Meta acquired Moltbook. The company lost its bid to acquire OpenClaw developer Peter Steinberger to rival OpenAI, and instead pursued Moltbook, the platform that Steinberger helped build. boring? perhaps. However, it continued to report news of Meta’s superintelligence laboratory.
