Flowith is useless.

"They're handing out gold bars in the village 🥵"

Haven't roasted an AI product in a while.

Finally, Flowith OS has arrived 🥺

My impression of Flowith has always been: it's a product only AI influencers use. Its convoluted interface and thoroughly mediocre output don't support normal user adoption. Can't make heads or tails of it.

But this thing keeps hyping itself up. To study the art of hype, I've tried using Flowith several times — and every time, I'm stunned by the interaction. Absolutely unhinged.

It's all just search boxes. I type a prompt, you spit out an answer. Done.

Flowith's brain does have some innovation, but it insists on flashy interaction gimmicks: showing you this interface where text boxes keep chaining together, desperately trying to look busy, and then the final response requires two manual clicks to open, which reveals — a sidebar.

This sidebar not only has tiny text, but can only expand to half the screen at most.

I don't get it (in Fenghua accent).

Can anyone explain why Flowith deliberately creates reading obstacles for users? What kind of — er, I mean, what kind of user-educating mentality is this?

Can't even be bothered to roast it.

But recently Flowith thinks it's hot shit again.

Released an "AI browser" aka Flowith OS.

Claims to have surpassed ChatGPT Atlas, to be the "world's first operating system purpose-built for Agents," and possibly "your last browser ever." Keeps chanting: world's first, surpasses Atlas.

Hype isn't the problem. Forced hype is.

I actually used it for two days. This thing is just an AI browser. How does that make it an operating system?

Flowith OS has two core capabilities. One, it can call a browser to help you post on Xiaohongshu, post on Twitter, organize web info. Two, it can call local terminal to help you write scripts, organize local folders, tweak terminal environments.

All fine and good. Just nothing special, nothing innovative.

Browser calling is table stakes for AI browsers. The social media posting use cases Flowith OS touts? Other AI browsers can do all of that too.

And I don't know why, but Flowith OS keeps emphasizing its ability to batch-write copy and post on social media. We generally call that behavior internet astroturfing.

The only place Flowith OS leads is that Atlas doesn't have local permissions and can't operate the local terminal.

But a month and a half ago, there was already a product that could do everything Flowith OS advertises.

StepFun Desktop Companion — can call browsers, can write scripts, can operate local terminals. More polished product, smoother task execution.

Flowith OS can't even come up with fresh narratives for its hype, and its product lags the market by half a year to a month.

The reason I don't use AI browsers is that I want to shop myself, post on social media myself. Beyond summarizing web pages and assisting with emails, I really can't think of other use cases.

But Flowith keeps cycling through the same scenarios, so I reluctantly tried to replicate one.

Everyone loves testing this thing with online shopping scenarios, right?

I tried it: asked Flowith OS to "go on Taobao and buy me a kitchen set, including a Chinese wok, soup pot, plus bowls, chopsticks, and plates. Prioritize simple, clean, Japanese-style designs. I cook at home regularly and host four to five people. Remember to claim coupons, add suitable items to cart but don't pay yet."

For this one task, Flowith OS ran for over 50 minutes. And beyond the initial QR code login to Taobao, it made me manually confirm things five or six times mid-process.

In the end it picked one wok and a two-bowl-two-chopstick-two-plate set. No soup pot. Didn't meet the "host four to five people" requirement.

Basically unusable. I'd rate this task's completion below 30%.

Second test task.

Flowith OS loves posting on Xiaohongshu, right?

I had it analyze all the copy from Flowith founder @dereknee's Xiaohongshu account and summarize a prompt that could one-to-one replicate his content style.

The result was completely wrong. Completely off-base.

Because it never actually entered Xiaohongshu. Hitting the login wall, it didn't ask me to log in manually — instead it web-searched "Xiaohongshu dereknee post content examples."

Then Flowith OS failed to scrape the web info, generated an empty file, and based on this wrong empty file, finished writing the Xiaohongshu account prompt.

Even more impressive: after this task completed, there's no way to interact. I realized its entire workflow was wrong, yet there was no input box for me to give new instructions. Task over means task over.

After seeing Flowith OS's workflow, I'm impressed.

Compared to this thing, Head AI at most qualifies as slightly overhyped marketing, My Shell is practically flawless in product completeness, and YouWare is as beautiful as the Venus de Milo.

Leon Ming, please come back. I'll never mock you again. You're the god of AI products, the paragon of honest, hardworking craftsmanship 😭

What can I say?

Looking at Flowith's hype history over the past half year, I only feel a kind of poverty. Poverty of creativity, poverty of courage, poverty of hype ability.

FDR said everyone should have four freedoms, one being freedom from want.

I fucking think these AI startup bros are deliberately antagonizing me, using their impoverished hype abilities to deprive me of my freedom from want 😠

First-rate AI applications create narratives. Second-rate AI applications can only steal narratives.

The biggest problem with AI applications is their lack of narrative ability. These startup bros/sis can't even cook up a new story.

Manus pioneered the "world's first general Agent" narrative, and other startup bros just swap out modifiers between "world's first" and "Agent."

Now Agent OS is hot, and these bros are swapping modifiers between "world's first" and "OS."

So after one or two years of grinding, everyone's capabilities are stuck at adjective innovation, never learning noun innovation.

No material, friends. Seriously no material — maybe just bite a lighter or something.

I won't even mention Flowith hyping infinite context a month ago, or calling itself "world's first infinite Agent" half a year ago.

I'm not asking for Genspark-level execution that can replicate features in two or three days. But you can't always trail industry hot trends by one or two months 😭

What's going on with this team? Not pursuing excellence hard enough? Can someone send a ByteDance colleague over to manage them, boost organizational efficiency?

Flowith's two domains combined: DAU just over 10K 🤓

Back to the title.

Flowith feels very much like Smartisan to me. You can't say it's useless, but it's genuinely not very useful. It has some design sensibility, but that sensibility comes at the cost of usability.

I still vividly remember in high school, a Smartisan fan classmate carefully showing off his Smartisan phone to me, like he had a stolen gold bar hidden in his jacket.

My first reaction to the Smartisan interface: this looks like Nokia, so retro in its flat design. Distinctive, sure, but not decisively different from other phones.

Flowith is also an idea-driven product. If the bro didn't tell big stories, I wouldn't say anything.

But in a very traditional, very mainstream category, relying on just one idea and some design sensibility innovations doesn't get you very far.

I struggle to imagine why people would abandon mature, stable Chrome; secondary-stable Atlas; tertiary-stable other AI browsers — for unstable Flowith OS.

One review on Guancha perfectly captures the user journey with Flowith OS: first opening it, nice design, looks good; actually using it, forced concepts, "using it is a pile of shit" 😆

Hard to keep a straight face

I also asked an anonymous Mr. Sun from Hegang whether Flowith is like a Smartisan phone.

As a former Smartisan fan, Mr. Sun reacted rather intensely:

Regarding Mr. Sun's unhinged remarks, we have no comment.

Objectively and rationally speaking, everyone's hyping operating systems, so I shouldn't only roast Flowith.

This company does have strengths. For instance, the guy can keep a straight face, shamelessly riding The Matrix by calling himself Neo, saying he'll "activate the Agent market like the chosen one" — had me dying.

Promo video's solid 👍

But, very unfortunately, at the actual product level.

Flowith has neither much innovation nor launched more than a month behind competitors. Everyone's talking big, and this bro took the big talk seriously. Built an AI browser, actually calls it an Agent OS.

This hype approach is a bit forced.

Manus, please come back 😭

When Man was here, I didn't realize how special it was; since Man went abroad, nothing compares 😭😭😭

Flowith can get valued at over $100 million. They really are handing out gold bars in the village 😭

(Article images generated by ChatGPT, had fun writing this, almost entirely manual writing 🤓)