Recent news comes out that Meta is acquiring Manus for approximately $2 billion, a move that appears to lack technical rationale.

What Manus really has is not some breakthrough AI. It is mostly a large amount of prompting and detailed instruction layered on top of the same commodity LLMs everyone else uses. They are good at building a fancy-looking UI that allows users who do not know how to code to run analyses and workflows. But most of this can already be done by tools like Claude Code or internal coding pipelines in a short amount of time by a senior engineer.

I do think the real strength of Manus is marketing. They hype themselves as top-tier AI, while in reality many strong AI engineers would not bother advertising something this technically straightforward. That makes it surprising that Mark Zuckerberg is willing to pay such a high premium to acquire it.

This also highlights a bigger issue: most incumbent companies keep saying they want to “do AI,” but they are moving extremely slowly. The real differentiator, as shown by products like Claude Code and Windsurf, is not the LLM itself. It is the deep understanding of the user’s workflow and how to design the prompting, orchestration, and backend logic so the AI fits naturally into that workflow. When done right, users are willing to pay because their productivity is easily 2–3× higher.

This makes me realize I need to put my head down and think much deeper about the “using” step in the chain — how users actually interact with AI day to day — and how to build an agent that makes the experience so smooth and effective that customers genuinely love it.