Clawdbot is a midlife-crisis hype campaign orchestrated by Elon Musk.

"The world is an information network"

Clawdbot is the most mid, most programmer-brained AI product I've ever seen.

This thing is identical to a NAS. Besides content creators and middle-aged electronics hobbyists, I can't imagine anyone needing Clawdbot.

How does anyone get hyped for a week over a phone remote control?

It's completely taking advantage of the fact that most people watching AI content creators can't deploy Clawdbot themselves and have no actual hands-on experience.

Clawdbot is just a phone remote. Its function can be explained in one sentence: use a mobile chat app to control an AI assistant running on your computer.

The AI assistant's capabilities come from having obtained system-level access to the computer, including local terminal, API keys, chat app account credentials — full permissions.

If you're bold enough to open up all sensitive permissions and let this thing grab your passwords, then sure, you can:

Send Clawdbot a message in your chat app and tell it to manipulate Claude Code running in your computer terminal to do work.

You can even infinite-loop it — you really can tell Clawdbot from your phone to use Claude Code to call a Skill to manipulate video editing software to cut clips, or manipulate Figma to do design...

An absolutely exquisite flavor of AI human centipede.

Clawdbot workflow diagram

Though I feel nothing but disgust for this Clawdbot hype that came out of nowhere, to properly roast this thing I still had to try it myself — otherwise I'd just be another big-company mid who only reads -1 reports 👋😭👋

Conveniently, all the cloud providers are chasing the new hot thing again, so I dropped 49 gold coins on Alibaba Cloud. (Later I'll explain how AWS hype in the US drove this whole 🦞 frenzy.) Shoutout here — Alibaba Cloud's pinned comment with the Wuying Cloud Computer deployment doc was way simpler and clearer than the main article.

It took me just over ten minutes to connect to DingTalk, so now I can @clawdbot in the Zang'ai Xianyu Tech Co., Ltd. DingTalk group and have it do small tasks.

Like checking the weather, writing a Gomoku mini-game, managing local files, or telling a Soviet joke.

I'm not sure if Clawdbot has restrictions built in, or if the Qwen-Max model it calls has unusually high moral standards. This thing won't do regional discrimination, won't do NSFW stuff, but Soviet jokes are fair game.

Kinda hard to judge.

If I had to name one strength, Clawdbot's memory function is decent. I asked it to call me Zang'ai Xianyu in every reply, and it did every time.

I can fully imagine that if I installed Claude Code on this cloud computer, or used my personal computer directly, Clawdbot could accomplish more tasks.

Like someone sharing how they're at the gym, sending a message telling Clawdbot to manipulate CC to write code.

Yeah, great. But why write code at the gym? Seriously, kinda sad 😭

I actually get why programmer friends find this exciting. Programmers' dopamine hits are different from normal people's.

Programmers just love connecting things together — like using their phone to control their home NAS, or using their phone to control their home Mac mini.

I can imagine the most satisfying thing a programmer could do right now: use their phone to control Clawdbot running on a Mac mini, hook it up with permissions for the whole Xiaomi smart home ecosystem, and use their phone to command Clawdbot to control small appliances.

Straight-up bliss overload.

I get it. The essence of a programmer is Brother Feng — they all love connecting.

But Clawdbot is completely useless for normal users.

First, why would I want to manipulate CC to write code from my phone? The screen is too small to read anything. The whole point of an IDE is to let me read and edit better.

Second, sure, there are some people who love working from their phone — plenty of bosses don't even use computers anymore. But they can use much safer, more mature solutions: tmux for remote terminal connection, and there are tons of bot projects that let you send messages from your phone to control Claude Code.

No need for this terrifying Clawdbot stunt of handing over local files and chat app credentials to a two-month-old open-source project that also uploads all your chat conversations to model provider servers.

Alternative solutions everywhere with a casual search

Most importantly, most people's FOMO here is incredibly abstract.

Claude Code is a real thing, and its adoption is still very low.

I tried teaching a non-technical Alibaba bro to use CC. After a week of tinkering, Google's Antigravity was still easier for him.

The Agent feature in Antigravity alone was enough to blow his mind — it could write actually usable work documents and investment strategies based on his local files.

The most certain opportunity in the next six months: the Claude Code wrapper competition.

Because CC is in fact the engine for all Agents. Put an easy-to-use shell on it, connect more useful API services, and you can create massive value.

And you'll find that on CC wrappers, Anthropic and startups are completely on the same starting line.

The evil Italian's own wrapper, Cowork, is terrible — tasks constantly interrupted, zero additional features. I deeply regret dropping 90 USD to try this thing.

Not as good as the CC wrapper my friend built in a month, proma.cool — at least tasks don't interrupt much, it's all calling the CC SDK, the actual work is no different.

Shell moats are real. When everyone's plugged into roughly the same engine, nailing the small features and interaction design of the shell to deliver a better user experience is a very certain opportunity.

We can list a whole roster of contestants: Jieyue Desktop Companion, Cherry Studio, Proma, MiniMax Agent — these are the ones I know, others please add in comments.

So I completely don't understand what the hype merchants on Twitter and WeChat Official Accounts are so excited about. Do you all just love connecting that much? 😭

That said, the Clawdbot hype frenzy isn't the developer's fault.

The Austrian bro open-sourced an AI assistant and didn't hurt anyone. And this T0-level hype wave couldn't have been personally engineered by him — if he had that skill level he should be working as Trump's special presidential advisor, AI would be child's play.

My independent thinking discovery is: the Clawdbot hype frenzy is mainly Elon Musk's fault.

Because Twitter recently updated its recommendation algorithm, which is supposedly fully LLM-driven.

This past month, my direct experience scrolling Twitter is: articles with millions of views, tens or hundreds of thousands of likes are increasingly common. These cross-circle super-viral hits keep emerging.

A typical example is Dan Koe — his writing is worse than Zhihu's top answerers, yet his motivational essays somehow penetrate Chinese and English-speaking circles alike, with a bunch of completely unrelated Chinese bloggers all posting their reflections.

A more direct example: the crypto circle is having a renaissance. Several crypto hype dogs are writing long Twitter threads sharing their trading insights. Some hype dogs have even emerged claiming they'll use media to do crypto investing.

The ones most suited to practice a16z's media-as-investment thesis are indeed the crypto hype dogs 😭

Hyped to oblivion

To support my observation, I used Claude Code to analyze Elon Musk's open-sourced X recommendation algorithm repo (https://github.com/xai-org/x-algorithm).

The results were tear-jerking — the core metric of the new recommendation algorithm is engagement.

This thing learns only one thing: what content makes users click, stay, and interact.

No content quality judgment, no clickbait detection, purely behavioral prediction machine. The model's scoring formula is brutally simple: final score = like probability × weight + reply probability × weight + retweet probability × weight + dwell time × weight ...

Meanwhile, the retrieval system mines "you might be interested in" content from the global corpus and pushes it to you, even from posters you don't follow at all.

These two mechanisms combined: high-engagement content naturally scores high + active cross-circle pushing = super-viral hits mass-produced.

Is anyone going to regulate Elon Musk? Twitter is literally hacking the entire world's population 😭

You need to imagine this world as an information network.

Twitter has the highest weight in the information network overall. Because a text-based community driven by recommendation algorithms, compared to short-video and subscription platforms, makes it far easier to create platform-wide viral hype.

Trump, Musk, and various KOLs are individual higher-weight nodes. NYT, WSJ and other traditional media were previously local area networks, now weakened into nodes.

Downstream from Twitter are TikTok, Reddit, WeChat Official Accounts, and even Douyin, Xiaohongshu. Because Twitter has higher weight, these lower-weight local networks will diffuse and propagate Twitter's viral content.

This means whoever hacks Twitter equals hacking the entire internet, able to pump shit into the brains of people worldwide.

And hacking Twitter only requires a dozen or two KOLs.

Both Google Search Trends and Github Star counts show that Clawdbot fully blew up on January 25th (Sunday), with a hype curve that could only be described as a vertical liftoff.

So what happened that weekend?

A group of English-speaking KOLs, mainly tech bloggers with a few crypto ones mixed in, collectively hyped "Mac Mini deployment," "five-minute AWS deployment," "first AI employee," "24/7 AI assistant"...

My head was spinning reading all these hype concepts.

After the weekend's English-speaking hype fest, Chinese-language bloggers on WeChat Official Accounts and Xiaohongshu quickly followed the trend — deployment tutorials, shock, explosion, I used Clawdbot for five/ten minutes to create insane cases, plus Mac Mini price hike memes flooded the entire internet.

However, while both Clawdbot and Manus triggered hype frenzies, their reputations differ because — with Manus you actually understand how to use it. A chat box, who can't use that? But Manus not giving invite codes so I couldn't try it made it pure hype, instantly identifiable.

With Clawdbot, most people genuinely don't know how to use it. Even following cloud providers' idiot-proof tutorials takes time to deploy and connect to chat apps. And how could people admit they can't use a trendy AI product?

Of course, there's plenty worth learning from this grand hype spectacle too.

Clawdbot's integration with chat apps is a传播 masterpiece on par with Manus's task execution visualization.

That pile of remote terminal connection tools — people genuinely can't tell what's impressive. But being able to @clawdbot directly in a chat app? Now everyone gets it, and you can even send voice commands.

Also shoutout to Tencent Cloud, Alibaba Cloud, Volcano Engine and other cloud providers for their quick reaction and solid hype game. One-click server installation is definitely easier than users typing commands line by line in terminal themselves.

Whatever this thing is, catching traffic is what platforms should do.

I seriously looked into it — AWS's official tutorial only caught up to the hype on January 28th, before that it was all KOL-driven. Domestic cloud providers actually reacted faster than AWS.

Also MiniMax, a model provider, started hyping Clawdbot as early as January 26th — this reaction speed makes me even more confident in Pangmao's B2C capabilities.

Finally, two more thoughts.

One: I genuinely think Elon Musk needs to be regulated.

He effectively owns the most influential media platform in human history, and has become the most influential person in this world.

It's hard to rank him against Trump, because Musk can keep hyping for decades, and he has absolute control over three tech giants: Tesla, xAI, and SpaceX. Among them, Twitter can genuinely hack the entire world.

Two: I sincerely suggest Qwen add a schedule assistant feature. User sends a voice message or chat screenshot, ChatBot sets a calendar reminder. Shouldn't be complicated, just call notification and calendar permissions — after all Macaron already implemented this.

Rather than letting every new AI product hype itself as a personal assistant, Personal Agent, might as well do it yourself directly. Push startup kids to innovate and revolutionize their hype game — big tech needs to give the kids some pressure.

That's all from me, basically done with work for the year, going skiing next week. Keep grinding, fam 🫡