How Manus found a $100M market in the ‘prosumer’ gap

This article summarizes an episode of  Xiaojun Podcast’s video series featuring Peak Ji, co-founder and chief scientist of Manus.

Yichao “Peak” Ji, co-founder and chief scientist of Manus / Photo credit: X

Yichao “Peak” Ji, co-founder and chief scientist of Manus, explains how the company went from a failed browser project to a company with US$100 million in yearly revenue by shutting down a finished product, which helped the team find a huge market of professional users.

The problem with selling new browsers

Before pivoting to agents, the team spent months building an AI web browser. The product worked, but users did not adopt it, and more development could not fix that.

Changing habits takes more than better features
Peak notes, “If you don’t think it’s cool, no one else will. Our team’s consensus was [that] the return on this might not be obvious.”

Big companies have a huge advantage
“Browser migrations in history… were driven by distribution,” Ji argues. “In an ecosystem where Chrome runs well and has excellent extensions, why would a user switch?”

Running AI on a laptop hurts the experience
Peak explains, “We found truly valuable AI tasks should be ‘long horizon tasks.’ But since you made a native AI Browser, the AI is running on your computer. I can’t close my laptop lid. So do I have to stare at it?”

Choosing survival over past efforts

The leaders realized that continuing the project was a mistake, not a sign of strength.

Releasing a bad product wastes time
Peak says, “If we released it, we’d be dragged into a cycle of maintaining it and self-justification, missing valuable new opportunities.”

Rival failures proved the decision right
“Josh Miller [CEO of The Browser Company] said: ‘I’ve been making Arc for so long, and I can’t even persuade my friends and family to switch from Chrome to Arc,'” Ji observes. “We thought, okay, browsers might truly not be suitable for startups to disrupt.”

It pays to admit mistakes early
Peak reflects, “This is my second time making a browser, and I reached the same conclusion… It is better to try quickly than to leave it hanging unresolved.”

Moving away from building everything

Leaving the browser market pushed the team toward a new plan. They realized that building their own AI models was a mistake.

Private models are copied quickly
Peak recalls, “In my last startup, proprietary models got eaten by unified models… The ‘Battle of 100 Models’ – 99 died already… the shelf life of a state-of-the-art model is only one to one and a half months.”

Speed and user data beat owning the model
The team realized that owning the “app layer” where users actually work is more valuable than owning the math behind the AI.

Finding the hidden expert user

Growing with general agents

The heavy structure for general agents

Confirming the main work tool