Meta has agreed to buy Manus, a Singapore-based startup that builds autonomous AI agents, in a deal announced on December 29. The acquisition brings Manus’s technology and engineering team into Meta’s growing AI organisation, while the startup will continue selling its subscription service out of Singapore. Financial terms were not shared.
Manus describes its product as a general-purpose AI agent, built to handle everything from research and software development to data analysis. Unlike familiar chatbots that wait for prompts and respond, these agents are meant to carry out intent. They take a goal, break it into steps, pick the right tools, do the work, and course-correct if something changes. In theory, at least.
Since launching in early 2025, Manus says it has seen heavy usage. The company points to large-scale processing activity and tens of millions of virtual computing environments spun up for users. It has also managed to build a real subscription business, which gives it one of the more tangible commercial stories among a flood of agent-focused startups that have appeared over the past year.
That kind of traction matters to Meta. The company has poured enormous resources into AI, from custom silicon to large language models, but turning that spending into products people actually use is still an open question. Buying a system that already has paying customers offers a quicker path to testing and rollout than trying to invent everything in-house.
By late 2025, Manus had reportedly reached a meaningful annualised revenue run rate, something that still sets it apart in a space where many startups are feeling their way through pricing and use cases. That momentum likely helped its case with Meta, which has faced growing pressure to show clearer returns on its AI bets.
Even so, strong sales do not resolve the bigger questions hanging over autonomous agents. Adoption varies widely by industry, and many organisations remain wary of letting software act on its own without close supervision. Issues around reliability, transparency, and how systems recover from mistakes continue to slow wider uptake, especially in complex or high-stakes settings.
There is also ongoing debate over how to properly measure these systems. While benchmarks exist for reasoning, tool use, and multi-step planning, critics say they often fall short of capturing how agents behave in the real world, where conditions are rarely clean or predictable. Manus has pointed to solid benchmark results, but as with much of the sector, independent validation remains thin.
The company’s roots are in China, where it attracted early attention after launching its agent technology in March 2025. As it expanded internationally, however, Manus shifted its strategic base to Singapore and scaled back parts of its China-based workforce.
That shift came as U.S. regulators increased scrutiny of investments tied to Chinese AI firms. One of Manus’s earlier funding rounds was reviewed under U.S. Treasury rules aimed at restricting certain capital flows into sensitive technologies. For AI startups operating at the cutting edge, where geopolitics and business are increasingly tangled, regulatory risk has become a real consideration.
Meta is hardly the only company chasing autonomous agents. OpenAI, Google, and Microsoft are all pushing systems that move beyond simple chat interfaces. The result is a crowded, fast-moving market, with big platforms often pulling smaller players into their orbit through deals or partnerships.
Still, the future of AI agents is far from settled. Supporters see them as powerful productivity boosters. Critics, on the other hand, argue that today’s systems are brittle, costly to operate, and capable of surprising, sometimes unwelcome behaviour. Many enterprises seem to agree, rolling agents out cautiously and keeping humans firmly in control, at least for now.
