The first week of March saw the release of Manus, a new general AI agent, from Tencent-backed startup Butterfly Effect. In contrast to DeepSeek, the agent uses a number of AI models, said to include Claude’s 3.5 Sonnet, and Alibaba’s Qwen. There is also a lengthy waiting list to get access. MIT Technology Review did get an invite code, and the writer described it as collaborating with an intelligent and efficient intern. The collaboration part is important and will include adjusting the prompt according to what kind of information it comes back with.

It also has a “Manus’s Computer” window so the user can see what it’s doing, like scrolling through websites when it’s searching for information. The most difficult task MIT’s writer Caiwei Chen set Manus was to compile a list of 50 people for the publication’s annual “Innovators Under 35” list, which she says is usually an “enormous” undertaking and involves sifting through hundreds of nominations. First, Manus broke the task into steps and settled on a search strategy that would include the widest possible selection of global candidates. Chen says it took three hours to produce a list of three candidates, but after repeated prompts, eventually came up with a list, although one skewed towards certain institutions and fields. Manus is also said to be capable of working asynchronously in the cloud, even when a user disconnects their device. Manus doesn’t work perfectly; some users report it getting stuck in loops, particularly if tasks aren’t defined clearly, reports DataCamp. MIT also said it couldn’t get access to paywalled content, but you can hardly blame it for that.