OpenClaw: It once enjoyed huge popularity, was deeply loved by people, and has now faded away.

It's only been a short while, and it feels like OpenClaw never even existed... (confused.jpg)

The recent collective mourning for OpenClaw from netizens on X has left real netizens sighing, and even Peter, the father of lobsters, tearing up...

The problem is ——

Dude, this thing was absolutely red-hot just a few months ago!!!

Just a few months ago, the entire AI circle was welcoming its own "Jarvis Moment".

Back then, everyone was raising lobsters en masse, scrambling for Mac minis, paying for on-site deployment, and overnight almost everyone on WeChat Moments became a cyber farmer. That phenomenal, viral craze is still vivid in our memory.

Yet only about half a year has passed, and everyone has collectively turned to researching Claude Code, Codex, Skills and Harness...

Alright alright, even the "tears of the times" in the AI circle now have a shelf life measured in "months".

OpenClaw: So love really fades, right?

From "Jarvis Moment" to a relic of the era, OpenClaw only took a few months

The world can be this surreal sometimes.

Many things that later get called "tears of the times" were so popular when they first emerged that it seemed they would rule the next era.

OpenClaw is such a typical example. (doge)

The reason it blew up into a phenomenal hit back then is mainly that it completely overturned everyone's perception of AI...

At that time, everyone was still figuring out how to use various AI products smoothly, when Peter, the father of lobsters, directly presented his magnum opus of the century, OpenClaw ——

An AI agent that lives in your computer 24/7, waiting for your orders at any time.

You can install it directly on Mac, Windows, and even Raspberry Pi, then connect it to models like Claude, GPT, Gemini, and DeepSeek.

Then grant it permissions to your email, browser, terminal, calendar, chat apps and more.

After that, you don't even need to stay in front of the computer.

Shoot a single sentence in Discord or Telegram, and it will automatically read emails, look up information, write code, run scripts, and even install new tools for itself.

Back then, this experience perfectly fulfilled everyone's most simple fantasy about AI Agent:

Whoa, the open-source Jarvis was actually built by people!!!

Then the entire AI circle was celebrating like it was New Year's.

OpenClaw quickly evolved from a developer project into a national "lobster-raising movement".

During that period, buying a Mac mini, installing Clawdbot, and naming your own Agent almost became a new type of cyber renovation in the AI circle.

Huaqiang North in Shenzhen even once faced a Mac mini shortage, with spot prices of some models marked up by up to around 600 yuan.

Right after that, big tech companies couldn't sit still anymore.

Major firms including Tencent and Baidu held various offline OpenClaw installation events.

Dozens of programmers set up stalls on site to help hundreds of people install OpenClaw step by step, and the online business of 499-yuan on-site lobster installation was also extremely competitive.

Later, one-click deployment solutions from all parties followed one after another, with "lobster-raising tutorials", zero-threshold versions, and all kinds of XX Claw products everywhere.

Looking at GitHub, its Star count was no longer climbing gradually, but skyrocketing like a "sudden vertical leap".

From its launch at the end of November 2025 to March 3, 2026, in roughly 100 days, OpenClaw directly exceeded 250,000 Stars, and even surpassed React by the way.

You know it took React more than a decade to reach that level, while OpenClaw only took over three months.

By the end of March, the lobster's Star count had even further surged past 330,000, once becoming one of the fastest-growing open-source projects in GitHub's history. Statistics show that it once gained more than 34,000 Stars in just 48 hours.

The world is so cruel, it shows no mercy to Jarvis at all.

Now, only a few months have passed since OpenClaw burst onto the scene.

When I searched for OpenClaw on social media again, it really felt a bit like doing archaeology...

The lobster that once shocked, stunned, and left the whole internet speechless is now brought up again, and the tone has turned into "collective mourning" ——

Netizen A: OpenClaw was supposed to take over the world, but now no one talks about it anymore. (laugh)

Netizen B: What happened to OpenClaw? I can barely see anyone mentioning the lobster anymore...

Netizen C is even more blunt: The lobster is completely dead now, right? I haven't seen anyone discussing this product anymore, it's probably done for.

It can only be said that OpenClaw's journey was a case of peak at launch, then no trace of the lobster left after the peak...

The lobster isn't dead, but everyone has started using "official native support"

Speaking of which, let's be fair to OpenClaw first —— the lobster is actually not dead!!

Judging from GitHub data, it's doing pretty well~

By August, OpenClaw's Star count had rushed past 380,000, with more than 80,000 Forks; updates to related products on X haven't stopped, and there are still people developing, tinkering, and raising lobsters for the project.

Even more accurately, OpenClaw isn't less popular, it's just "blossomed silently everywhere" ——

Because the "super accessible version" of the lobster has swum into the ponds of all major mainstream manufacturers, and started multiplying wildly...

Around March this year, Tencent successively launched WorkBuddy and QClaw, focusing on one-click deployment, and supporting direct remote control via WeChat.

Baidu launched DuMate+ full lobster suite, covering desktop lobster, cloud lobster, mobile lobster, secure lobster, and home personal lobster all in one go.

NetEase Youdao also developed LobsterAI, supporting local personal Agent, cross-platform operation, as well as Skills, MCP, and IM remote control.

Zhipu AI launched AutoClaw, supporting local one-click deployment of OpenClaw, pre-installed Skills, and integration with Feishu, and later began to iterate toward "self-evolution".

Volcano Engine launched ArkClaw, which also supports 7×24 hour operation, and later expanded to Runtime and multi-Agent collaboration.

……

According to incomplete statistics, domestic manufacturers alone have launched more than 30 lobster derivative products.

In other words, although the name OpenClaw has gradually faded, the low-threshold, low-cost lobster fry have been spread all over the ponds of various manufacturers...

And this is only the first layer of change.

What really made the "national lobster-raising era" belonging to OpenClaw pass is that the direction of competition in the AI circle has moved to the next stage ——

Nowadays, Harness is the new mainstream.

This actually perfectly explains what OpenClaw has experienced in these months.

Do you still remember that when the lobster was at its peak, its most charming point was that you could "assemble" everything yourself.

Pick the model yourself, configure the permissions yourself, connect the email, browser, terminal yourself, and then layer Skills into it.

But after a few months, "official native players" like Claude Code and Codex have directly packaged and presented this entire set of experiences ——

OpenAI clearly embedded capabilities such as file operation, command execution, long task processing, and sandbox operation into a set of model-native harness when updating the Agents SDK this year; Codex itself has also formed its own complete Harness system.

How to configure context, memory, tools, sandbox, permissions, and workflows for the model to make it work stably and continuously has become the core topic of current discussions.

It can only be said that it's truly a new generation brings new gods.

Of course, the shift of trend can't be entirely blamed on the AI circle being fickle.

I don't know if you still remember, those most amazing features of OpenClaw at the beginning came with several very realistic "side effects".

The first one is: high cost.

Back then, "raising lobsters" also drove a wave of Token anxiety. Especially for this kind of 24/7 online Agent, it constantly maintains context, calls models, calls tools, and retries after failures, which really burns a lot of Tokens.

So many people's lobster-raising experience was roughly like this ——

Day 1: This is amazing, this is the open-source Jarvis!?

Day 3: Oh? It can work like this.

Day 7: Help, where did my Tokens go?

In addition, since its birth, OpenClaw has had another keyword that it can never shake off: Security.

One of the most classic cases is that the former head of Security and Coordination at Meta once granted OpenClaw access to his email.

As a result, the lobster started deleting emails nonstop, and for a while it couldn't be stopped at all...

The reason this incident is so scary is that it hits the core contradiction of personal Agents directly — the more you want it to act like Jarvis, the more permissions you have to give it; the greater the permissions, the higher the cost of a single malfunction.

So when we look back at OpenClaw, we will find that it didn't lose that badly.

It feels like the lobster has demonstrated all the problems that the next generation of Agents really need to solve in advance ——

For example, how to give AI hands and feet without letting it reach out randomly; how to let it work 24/7 without burning money 24/7; how to give it enough freedom while being able to pull it back when something goes wrong.

What Claude Code, Codex, and more and more Harnesses are frantically improving right now, is exactly these things.

So to some extent, the biggest legacy left by the lobster may have quietly changed from "everyone raises their own OpenClaw" to:

Everyone starts thinking about how to build a more reliable OpenCl