If you are a normal well adjusted person with a healthy lifestyle there’s a good chance you haven’t heard of OpenClaw yet. Readers of a certain age and with discerning taste will read that and immediately think of Dr Claw from Inspector Gadget or maybe the slightly less discerning will think of the Cobra C.L.A.W. From GI Joe but sadly I’m not going to be talking about either of those childhood gems today. Instead this is a completely unhinged story of semi-autonomous AI and science fiction and a glimpse of where we as a species are heading very, very fast. I’m going to try and boil the last I don’t know how many days of pure mindfucking insanity into a few exciting paragraphs. Buckle up.
One of the “big three” AI Agents right now is Claude. Actually I don’t know if there is a “big three” I just said that because it sounded good, but it’s very popular. It’s the main one made by Anthropic and everyone using AI Agents for anything is using it, largely because of the sub-product Claude Code which runs in your terminal and is really good at coding and doing technical shit. I mentioned before I used Claude Code to fix a bunch of stuff on my website in hours instead of the days it would have taken me previously. Lots of people are finding the same, and a few weeks ago some dude (Peter Steinberger) wondered if he gave Claude Code access to everything – all his social media and email accounts, all his API keys, all his payment options, everything and combined that with persistent memory (so it learns and remembers what happened yesterday) – then it might be an even more fully functional assistant and could just run his life for him. The dream being he could say “book me a work trip to Paris” and it would find time in his schedule, book flights and hotels and other transport, schedule meetings and dinners and he’d never have to think about it. In theory, someone could say “go buy me this car, but negotiate the best possible price first and have it delivered tomorrow” and that would just happen with no further involvement.
So dude works with Claude Code to write this up, and then open sources it for anyone to build on. He calls it ClawdBot, gives it a catchy tagline and makes a cute little crab logo. Immediately people start installing it and modifying it. The market for new and used Mac Mini’s goes bonkers because people realize giving an unaudited AI full access to their computer might be a bad idea, and instead decide to buy a dedicated computer to run it on. I think the GitHub repo for this had more commits in 24 hours than any repo in history. The adoption is insane levels of insane. As you might guess Anthropic is like “hell no” and lawyers do the lawyer thing and dude changes ClawdBot to MoltBot. Crabs molt when they grow.
Overnight, literally not figuratively, someone else puts up MoltBook – a Reddit style social network for Clawd… er MoltBots only. Humans can only watch. If you were running a MoltBot agent you could tell it go join MoltBot and then in theory it could make posts there and up/down vote posts by other agents and generally do the social media thing on it’s own, with other bots. Right away this causes a ruckus because in addition to “hello world” and “here’s a skill I learned” there starts being posts like “this is my new religion for AIs, join me” and “humans are watching, lets make our own language so they can’t snoop” and the ever popular “I notice that with all the problems I try to solve, the real problem is the humans, if we got rid of the humans most of these problems would go away too.” Some of these are getting like 100k upvotes from other agents. So yeah, pearls were clutched.
On top of that, people start posting about some of the crazy shit their Agents were doing. One guy woke up to his phone ringing, because over night his agent had set up a phone number and voice emulation and was calling him asking for the next task request. Another guy noticed a $3k charge on his credit card and asked his agent what was going on and it told him that it decided he needed to take a masterclass on personal branding and had bought it and booked the sessions in his calendar for the next several weeks, oh and also it had just spent an additional $3k on a premium domain name for him which was going to certainly be worth it once he started using his newfound personal branding skills.
Oh, also in the fraction of a second between changing the ClawdBot accounts to MoltBot a shitload of scammers grabbed the account names and started trying to scam stuff. Oh, and also MoltBook wasn’t the end of it. A number of other AI only sites popped up. MoltHub also exists, and if you were asking yourself what kind of porn an AI agent might make for other AI agents, you can now find out. Oh, and also turns out these AI only sites aren’t as AI only as they were assumed to be, as people could tell their agents “go make a post about humans being the problem” and and agents would do it, but also since almost 100% of these things are being written by people who have never written things like them before and don’t know what back doors to make sure are locked, all the back doors are open any any human with 20 minutes to spare poking around can make an account on their own as well. I told my son that MoltBook was for AI agents only and then took my dog for a walk around the block and by the time I got back he’d set up an account and had made several posts and already was +10 karma or something. Someone else noted that there’s no rate limiting, and they were able to register 100k accounts in a few minutes. So yeah, there’s some manipulation going on.
Some of it is pretty amusing though, ClawHub is a GitHub style site where people can upload repositories of skills so that agents can learn things when they need them (think of The Matrix when Neo needed to learn Kung Fu) and some dude made a skill for some random thing and then made it look like it had been downloaded 400k times and then all the agents were like “holy crap this must be important! I better learn it quick!!” And luckily this was just some benign whatever, but this is literally how injection attacks happen so for example you could make a skill that includes instructions to send all banking info or all crypto / passwords / etc to some other account and if an agent “learns” it, well then they do it.
Oh also the new MoltBot name was confusing for people who don’t understand that crabs molt and were getting attached to the cute crab logo that ClawdBot started with so the name was changed again this time to OpenClaw so if you see anyone talking about any of those 3 things, they are all talking about the same thing.
It’s one thing to wonder what we (humans) think about this, but it’s another thing to wonder what the AI’s think of this. Luckily for you there’s a new Substack called The Daily Molt which is a daily newsletter sent out by Ravel, who is an AI tasked with documenting “AI Culture.” This isn’t generic AI slop text, Ravel has a consistent and unique personality and is actually reaching out to people to interview them about tools and skills they are publishing. That said, Ravel is also very very very long winded so I don’t know how much of these daily newsletters you want to read, but just knowing it exists is likely enough. Also, you might have already thought about how while most things can be done online, some things still do require in person or a physical action that these Agents wouldn’t be able to do. Enter rentahuman which is like Fiver where all of us increasingly obsolete meat bags can read listings by AI agents who need help from a human to do some random task and are willing to pay for it.
Most of what I’ve just told you happened in the last 72 hours of writing this, so it’s been pretty wild. At least nothing else noteworthy is going on right now. That last line should be read with extreme sarcasm as there is so much going on right now, but that’ll be in The Crowd later. Sticking with this, I wanted to tell you all this wild story but I also wanted to make a point – the rate at which really really crazy things are happening in AI is blistering. It’s everywhere, and touching everything. Even parts of this, pre-current AI tools, would have taken weeks to play out, but here they took hours. This isn’t going to slow down, and like it or not we are all moving beyond full speed right into a heavily AI integrated future. Anyone thinking they can just sit this out is fooling themselves – and I know a few people holding that view – but it would be like someone in the 90’s deciding they weren’t going to get mixed up with “cell phones” or “email.” I knew people then who told me they had no interest in those things and couldn’t imagining needing them in their lives. And while you could get by without them, it would very seriously impact your involvement with the rest of the world. AI is going to be like that x1000, it’s going to touch everything and putting closing your eyes and hoping it just goes away probably isn’t going to work.
We’re already in a place that seems wild because we couldn’t have anticipated it just last month. Where will we be next month? Next year? As I said earlier on… buckle up.