Manus remains the flavor-of-the-patch god.

"Wrapping Is Hard"

"The most important thing about being online isn't learning — it's having fun ❤️

My fellow AI entrepreneurs are running out of material lately. In light of this, I'm launching the Big Picture series, offering strategic takes on Agent platforms from the boss's perspective.

Our motto: Everyone becomes Xing Wang, we all get to make a splash.

Episode one: Manus. Episode two: Coze Space. Episode three: MuleRun."

Today we're playing the long game.

From Manus to Macaron, a lot of people feel this wave of Agent hype has peaked, that there's nothing left.

New Agent products still launch daily. But none of them are fun. Either they're B2B components — boring — or they copy Manus's approach with even jankier frontends.

The most interesting one is actually Macaron. Whatever you think of the product, at least the guy's got the guts to tell a story. Spending VC money to entertain the masses — isn't that the noble spirit of the macaron? 👊🤣🔥

You can blame all of this on Manus. (Damn Singaporeans, why they gotta be like this 😭)

Over the past six months, I've only used Manus twice since that first launch month.

The most recent time was to find new maps for Red Alert — I'd gotten tired of the old ones. Manus found me a Baidu Netdisk link with playable new maps. Problem solved. But apart from these extremely niche use cases, I genuinely can't think of what else to use Manus for.

The core dilemma facing general-purpose Agent products is insufficient user demand — people simply don't know what to do with all these Agents.

I checked Manus's own recommended user cases. Fifty cases on the homepage. Half are search scenarios, writing research reports. The other half are generating web pages and mini-games. Overlapping between them, a dozen-plus cases involve writing a report first, then turning it into a web page or PPT.

https://manus.im/usecase-from-user

What does this tell us?

It tells us general-purpose Agents don't work. General-purpose Agents aren't general-purpose — their main use cases boil down to two: writing reports and generating mini-games.

Narrow it down further, and all general-purpose Agents essentially offer two functions: DeepResearch and 4399-style mini-games.

Neither of these scenarios, standing alone, is particularly valuable.

DeepResearch's biggest problem: nobody can get through an AI-generated report running thousands of words. We can't even be bothered to read human-written reports most of the time — it's not like we'll suddenly fall in love with learning and devour AI-generated literature reviews.

AI-generated long reports have neither original insights nor primary-source information. And for pure search capability, top-tier models' end-to-end search — GPT-o3, GPT-5 Thinking — is already sufficient.

Generating web pages and mini-games has the same problem. Nobody looks at AI-generated frontends. This publication has repeatedly argued that product definition is an extremely scarce capability.

AI-generated mini-games will never be as fun as 4399's catalog. Because even the simplest game requires genuine thought and creativity — not something you can fake by feeding a few lines to an AI.

But here's where Manus creates value: by combining these two functions.

An AI-generated report of several thousand words? Nobody's reading that. But package that same report into an interactive web page — still no external audience, but the user themselves can actually engage with it.

Take one Manus case: studying Geometric Unity theory. It transformed a theoretical review into a web page packed with AI-generated diagrams, embedded YouTube explainer videos.

https://manus.im/share/OIEfSzUqePyLH70UXbjFPH?replay=1

Isn't this a new media format?

Because its supply side has fundamentally changed. AI-generated text is cheap; generating frontend web pages is trivially easy. The cost of producing this multimodal, aka flashy knowledge-webpage format, has dropped to essentially zero.

The value isn't in the AI-generated content itself. The value is in packaging that cheap AI-generated content into an appropriate media container.

This is indeed wrapping. And wrapping is genuinely difficult. Manus created a wrapper, and actual users are willing to pay for it — that's innovation in an absolute sense.

So I still consider Manus the GOAT of this era.

Manus pioneered the general-purpose Agent era. Early days, sure, but still the pioneer.

First, product form.

A large model on the left handling task orchestration, a cloud computer sandbox on the right for the model to operate within. This product form was defined by Manus.

Half a year later, every general-purpose Agent on the market — Genspark, Coze Space — is still following Manus. Not one has produced innovation that overturns this paradigm.

Then, narrative capability.

A Chinese guy, sitting on a WeWork lobby sofa, explaining product philosophy in English. That launch format was invented by Manus.

The startups that came after couldn't even come up with new talking points for their promotional videos. Just noun substitution, swapping out modifiers between "world's first" and "Agent." How can you not miss Manus's storytelling 😭

Here I must praise Macaron again — the guy used two camera angles in his promo: a tracking shot from the side, and a front angle that actually zooms. Introducing the zoom concept to AI product promos for the first time — now that's narrative innovation 👍

But beyond this, has any Agent product produced innovation that surpasses Manus?

No.

Coze Space has impressively realistic AI podcast voices. Genspark is endlessly iterating on PPTs. MiniMax Agent beefed up vibe coding capabilities, seemingly gunning for YouWare's business 🤓

But these are all innovations within the product form that Manus defined — none have created a next-generation product.

Especially Genspark. Originally an AI search product, it hyped itself into relevance by copying Manus as closely as possible, claiming $36 million ARR just 45 days post-launch.

The founder constantly appears in promos with dark circles under his eyes. I'm afraid to even use Genspark — every click of the send button feels like I'm draining the man's life force.

Manus, meanwhile, for all its hype-driven origins, only announced $90 million ARR six months after launch. Compared to the circus of hype these past six months, that deserves to be called steady and grounded.

Everyone thought Manus was a transitional product. But six months on, nobody has surpassed it.

This proves Manus is indeed the GOAT.

Well-read friends know this pattern has repeated throughout internet history.

Browsers were initially dismissed as transitional. Same with PDFs. Many felt these were too limited, not technically sophisticated enough — but they were simply too easy to use in certain respects.

One unmistakable trend these past six months: general-purpose Agent products have a real shot at replacing the office productivity suite.

General-purpose Agents may fail at everything else, but they can genuinely accomplish entry-level white-collar work — writing reports, making PPTs, crunching spreadsheets. This represents a platform-level application opportunity.

The only question is whether the AI office suite opportunity belongs to online collaboration platforms like Lark and Notion, or to standalone Agent products like Manus and Coze Space.

That's something we'll only know when they clash head-on.

(Illustrations generated by ChatGPT o3, with writing assistance from Gemini 2.5 Pro. Made a video version 👇)