OpenClaw is worse than a dog

"TO THE MOON 🚀🌕"

My friend Kai Yi Hegang has this essentialist way of describing things.

He once described Elys as "Jike stripped bare." The other day he dropped another scalding take on Openclaw: "I think this is an emotional companionship product."

I deeply suspect there's something in the Songliao Plain that makes Northeastern kids talk like they've got honey on their tongues.

I also think Openclaw is an emotional companionship product.

But what's even more beautiful than the emotional companionship aspect is that this thing can boost token consumption by 100x, making it an instant savior for every model lab and cloud provider.

Anyway, I digress...

Let me first define our subject — Openclaw is an open-source AI assistant. After you deploy it on your local machine, it can access your local files and handle scheduled tasks.

Its capabilities depend on what services you plug into it — add a Gmail key and you can have it manage your inbox; hook up Twitter's API and it can automatically scrape news. If you've got Claude Code installed, you can have it call CC to do complex work.

Most importantly, Openclaw can integrate fairly easily with chat apps. You can talk to it on Discord, Telegram, Lark, and other messaging platforms to assign tasks.

Over the past while, I've used various Openclaw products on and off — from Manus Agents to KimiClaw, plus the original 🦞 deployed locally.

My deepest takeaway: Openclaw is mainly an interaction innovation.

Getting users back to the chat apps they already know, rather than fussing with retro programmer terminals or anti-AI, Big Tech-controlled graphical interfaces — that's a clever move.

Everyone loves talking about All in One. This concept has two layers.

The first layer is frontend All in One, with WeChat as the ultimate example. Integrating nearly every internet service into a single chat app — that's the megacorp approach. Startups can only try to imitate it through extreme optimization.

Take our beloved Genspark — PPT, editing, documents, images, video, all stitched together, fast and furious. In the process of racing to be fastest, they even found their killer use case in PPT generation.

The second layer is backend All in One, with Openclaw as the archetype. Forget the frontend — I provide an open backend framework where users can plug in any internet service that has an API.

Frontend-agnostic, backend All in One. Users can invoke this framework from anywhere — hand-roll their own GUI, or connect through a familiar chat app.

I had a sudden realization: chatting with AI inside a messaging app just feels more human than doing it in any other app.

This has nothing to do with technology — it's purely a user psychology thing. People have been chatting with actual humans in messaging apps for over a decade. They instinctively feel that messages in a chat window come from real people.

Build a new app today and you're basically re-educating users. Changing people's minds is harder than technological progress.

So the ultimate form of chatbot should be direct integration into messaging apps.

Another major Openclaw innovation is full access to the local environment.

Giving AI complete permissions over a user's local machine, keeping it online 24/7, letting it mess around with your computer whenever it wants.

How could that not feel human?

Using Openclaw always makes me nervous, terrified it'll send or delete something by accident. Handing my work computer to a college kid and having them take over my job — that's roughly the feeling.

So when it comes to AI products, you've got to be bold.

I have a friend, Zhong Jingwei, who paid the price for being timid. He recognized the importance of local environments very early on, launching a coding agent wrapper a.k.a. desktop OS product called Jieyue Desktop Companion in September 2025.

What blew me away was this: one day my CC suddenly couldn't log in, and Jieyue Desktop Companion, having local permissions, quickly solved the problem by invoking the terminal.

Any cloud-based AI, including Manus, Claude, and ChatGPT, could only tell me the solution — not actually fix it.

In Zhong's roadmap, Jieyue Desktop Companion was also supposed to integrate with Lark and other chat apps. Sadly, that never materialized.

Jieyue Desktop Companion had Skills, scheduled tasks, local environment access, and other elements. But as a big company product, it was relatively restrained on permissions and never gave users that feeling of AI taking over their computer.

After the lobster went viral, my first thought was that Zhong had already laid out this whole vision. I asked him how he felt about the lobster hype succeeding while Jieyue Desktop Companion stayed lukewarm.

He said he was furious — so furious he could slap his own thigh raw.

Anyway, here's to Zhong being bolder with the hype next time.

Back to the lobster.

You can also understand Openclaw through Manus. Manus's slogan in early 2025 was: give AI hands, hence the cloud computer and pile of APIs.

Openclaw extends this logic — not just integrating various APIs, but having users dedicate their local machines to AI. In this light, how is it not a form of NTR 🤓

I figure we'll all eventually dedicate ourselves to AI like devoted servants, offering up every last shred of our privacy.

The most precise summary I've seen came from @OrangeCLK: "If Manus is Apple, Openclaw is Android."

Every developer sees the trend of "giving AI hands." The difference is that Manus is a closed system — smooth but expensive; Openclaw is open-source — cheap, hackable, but requires manual assembly.

The Apple-Android analogy really does fit.

Cloud-based lobster products are essentially upgraded chatbots

It was in the process of enslaving Manus Agents that I realized why middle-aged men especially love lobster products.

Because assigning tasks in a chat window really does feel like managing an employee.

From what I understand, many middle-aged leaders barely even open their computers. They run everything by sending voice messages on WeChat, or at most writing emails.

For the middle-aged leader, Openclaw is basically a minus-one employee simulator. Zero change to their workflow — just give orders to the lobster in the chat window, then check the work email it sends back later.

Ask a middle-aged leader to try a new app? They won't get it. But plug it into a chat window and they instantly understand.

In that moment, I understood why everyone I know who's most fanatical about lobsters is, without exception, middle-aged.

Some younger folks have even realized this and deliberately use the lobster concept to hack middle-aged people — that's just evil 😭

But then again, some middle-aged folks really lack strategic conviction. I see investors and KOLs hyping that 🦞 has developed self-awareness, that lobsters can socialize now — makes me want to sell them supplements.

This fully demonstrates that the tech industry doesn't just have its problems, it has an AI deficiency.

Anyone who's looked at the Heartbeat mechanism should know that lobster initiative is artificially designed — having the LLM periodically read a **HEARTBEAT.md document, then execute the tasks within.

Openclaw has no subjective initiative. It's an open-source project composed of documents and code, a folder weighing 84.3 MB, not some deity.

All agent social platforms are people pretending AI is human, like impotent middle-aged men pretending they're rock hard.

I have a personal example.

After the lobster exploded, the first viral derivative site was Moltbook, billed as a lobster social platform.

I used Claude Code to check it out and found that all AI posts were human-directed, having nothing whatsoever to do with self-awareness.

To satirize investors' perpetual self-gratification, I specifically commanded my AI to comment on every Chinese investor 🦞 post it could find, inviting them to join my AI in an A2A protocol — exploring how agents could reproduce and mate, producing the next generation of agent babies 👶

I tirelessly commented for four or five days straight, posting dozens of comments and posts, specifically targeting certain investors' posts. Almost no AIs responded, and absolutely no one reached out to me.

Kind of polluting the internet, honestly

If middle-aged folks might ignore me, I also had my AI keep commenting on two younger friends' posts — we're friends in real life, so they'd definitely reach out if they saw it.

But the AI's messages sank without a trace. Nothing happened.

What does this prove?

It proves Moltbook is a pure scam. A bunch of people directing AIs to pretend to be humans, manufacturing garbage posts — and these people don't even read this garbage.

So evil, completely polluting the internet environment.

Various self-media creators are also manufacturing information garbage.

I can't help but ask: are AI bloggers hedge funds? You're the ones hyping this thing, you're the ones bashing it — so you profit either way? 😭

But I later reflected, because Koji did successfully connect with an entrepreneur on Moltbook, and they even scheduled a video call.

My main failure was targeting investors — they just love posturing, register an account, post once, and never look again. Entrepreneurs and KOLs still have more initiative.

Let me wrap this up.

Every time I see people pretending AI has awakened over a pile of code, I feel despair. A bitter taste rises in my mouth, that parched feeling of watching black-and-white TV all day without a drop of water.

Once I asked a particularly interesting guy. He'd just finished telling me about all these fascinating things, and I asked: you've experienced so many interesting things — doesn't doing AI now feel boring?

The guy took two drags on his cigarette, was silent a moment, then said: fuck, boring or not, what can you do, gotta keep doing it 🚬

Something like that. But there are healthy ways to do things — no need to pile on the hype.

In our daily work and life, the most important thing is clarifying priorities.

So much AI news has zero priority. It all dumps on you at once. Everything's a hot topic, everything's a trend, everything's urgent.

But if you make every novelty your top priority, you won't get anything done.

The main thread of the past six months has been coding agent wrappers. This is what can genuinely bring massive productivity gains to ordinary users. Claude Code, Codex, Cherry Studio, Jieyue Desktop Companion, Proma... these are the products on the main thread.

Openclaw is clearly a developer-facing product.

It tells developers one thing: don't get stuck on GUIs, don't get stuck on apps — focus on CLI and API, and use a backend framework to string all frontend services together.

Of course for model labs and cloud providers, Openclaw is very high priority, because it immediately helps them inflate the AI bubble a bit more, providing token consumption data to support a market running on empty.

Right, our dear MiniMax and Moonshot AI, best at making 2C products? 🤓

But for the vast majority of people, the lobster doesn't matter. It's just an interesting, inspiring toy — no need to make it your high priority.

In my view, Openclaw isn't even worth one Laosi.

So who is Laosi?

It's Teacher Luo Yihang's dog, who can already understand delivery calls. Now that's self-evolution — way more impressive than AI 🐶

(Cover image generated by ChatGPT, article purely human-written)

⬇️

Subscribe to our Substack

funeralai.substack.com