About the overnight sensation Manus, I don't really understand it myself. I'm reposting an analysis article from AIGCLINK on Twitter for everyone's reference:

———-

Last night, the Chinese-speaking internet was flooded with PR about Manus, claiming it to be a universal Agent. Late at night, I had a chat with Teacher Guo from Baichuan and a partner from MGX, and here are a few insights to share:

1. Breaking down Manus: Its capabilities are a combination of compute use + virtual machine + artifacts + a built-in set of agents.

2. Feasibility: If it doesn't get internalized by models, an agent is a personalized thing. Manus seems to aim for a universal version of bolt.new (which focuses on programming tasks), which is clearly contradictory and unlikely to succeed.

3. Possibility: The only viable path is positioning it as a new entry point, integrating various agents and compute use capabilities in the future. This would be an enormous workload. A more promising approach would be aggregating via the MCP protocol model—unless they take this route.

4. Market positioning: It might explore a new universal model like AI search, but with a higher barrier to entry. It may not become something the general public can quickly adopt.

5. Limitations: Those who understand it might disdain using it due to its constraints—unless there are a few breakout use cases. Those who don’t understand it won’t use it, but it could become trendy among content creators who like to show off.

6. Competitor 1: If it truly is a universal agent as advertised, these universal capabilities will likely be internalized by large models in the future.

7. Competitor 2: The main industry competitors are coze and dify. If this path proves viable, they’ll likely adopt it too.

$NVIDIA(NVDA.US)$Tesla(TSLA.US)$Taiwan Semiconductor(TSM.US)