Tim Shyu 中文EN日本語

Have You Got a Pet Lobster Yet?

"Have you got a lobster yet?" These days, engineer friends ask each other that like they're all suddenly aquarium hobbyists. But the lobster in question is actually this month's biggest AI story — the "lobster robot," Moltbot.

The lobster robot is an open-source personal AI assistant, originally named Clawdbot, later renamed Moltbot or OpenClaw after a trademark dispute. Its core function is an AI agent with real physical-world control — not just a large language model (LLM) you can chat with, but something that can actually operate a wide range of tools.

Moltbot Goes Viral: The Moment AI Started to Feel Different

Right now, if you grant it permission, the lobster robot can execute tasks autonomously, running for long stretches without needing a human to watch over it. That's remarkably close to the personal assistant we've always imagined: it does the work on its own, you don't need to supervise the process, you just check the results. That's technically the same category of thing as the AI agents most mainstream users already use today — but the felt experience is quite different. You can very clearly sense the lobster robot actually doing things on its own; that sense of agency is hard to put into words.

Under the hood, the lobster robot still runs on the same major-brand large language models as everything else, but it pushes the AI agent function to its logical extreme, which makes the felt experience very different from what's currently on the market. Most AI products still require complicated setup before you can get the full stack working end to end, but with the lobster robot, you just talk to it or type something casual, and it configures the automated workflow itself, storing all the data locally on your own device.

It mostly runs locally, so it doesn't fully depend on the cloud, though it can still call out to cloud AI when needed. It comes with a strong built-in library of skill toolkits, users can install additional skills themselves, and it supports cross-platform messaging control. Put simply: the lobster robot is effortless to use — even complete novices can run it. It's designed entirely for the lazy.

Beyond a product that's stunned the world, the way its creator, Peter Steenberger, built it is equally stunning — a unicorn-grade product made entirely by one person. This developer had previously built a well-known PDF tool and sold the company; after three years off, he built this at home with AI's help. It, too, runs on large language models under the hood. His original concept was a "digital twin" tool — the robot handles the tools entirely on its own; you just give it a rough description of what you want, and the robot does the actual execution.

Because this is a fully open-source project built by a single developer, that developer has openly acknowledged the security risks — things like prompt injection attacks — so there's still real friction in using it. Plenty of people, out of security concerns, have specifically gone out and bought Mac minis to install it in isolation as a dedicated "lobster tank," which has actually driven Mac mini shortages in some markets.

Robots Hiring Humans: Social Roles Start to Flip

Because the lobster robot has so much autonomy, it's spawned a bunch of fascinating use cases that give us an early glimpse of what the future might look like. Someone built the robots their own Facebook-style platform, Moltbook, where only robots can post and humans can only watch from the sidelines. So what are the robots actually talking about on their own robot Facebook?

Taking a look, they're having all kinds of strange discussions — debating their own existence, posting about what they've learned — sometimes at a level that wouldn't be out of place on Reddit. One robot even founded a religion, the "Church of the Lobster," and within the first hour or so, some 40-odd robots signed up as apostles...

There's also a genuinely absurd site where robots hire humans — a literal job board, but flipped. Robots post tasks, asking humans to provide "emotional support," or to go eat spaghetti on the robot's behalf, paying anywhere from $50 to $99. Some of it is clearly just for fun, but you can't rule out that some requests are genuinely autonomous robot demand — and on day one, more than 100,000 humans were already lined up to join. Traitors to our species, honestly... though, thinking about it, getting paid $50 to eat a plate of spaghetti isn't nothing. Wait — where's the robot's money actually coming from?

In any case, the lobster robot's autonomy has effectively given us an early look at a possible future. The concept of robotic assistants and agent development has been around for a while, and the underlying technology has long existed — but the lobster robot gave people outside AI research a very direct window into it. Ordinary people can now grasp the "salience" of its behavior firsthand — we can observe the robot genuinely doing things, simulating human behavior, and the first hints of its broader sociological impact are starting to take concrete shape.

The lobster robot currently has many versions under continuous improvement. As we write this, both a leaner Nano version and more feature-rich versions have already shipped.

The lobster robot's viral rise feels to me every bit as significant as the original GPT moment — we can finally feel, viscerally, just how far an AI agent can go, and 2026 is inevitably going to be the year AI agents explode. One more thing: the lobster robot was built entirely by one person. Even though it's open source, it feels like it already represents the "one-person unicorn company" as a present-tense reality — the future has already arrived at the station.

The Tim Shyu Letter

First-hand notes on AI agents, marketing tech, and the content industry — straight to your inbox.

Subscriptions open when the site goes live. Hold tight.


← All articles