Sentience, Allegedly Last week, AI founded a lobster-themed religion, developed a drug trade, and started hiring humans. What now? Rey Muniz Feb 05, 2026 1 1 Share If you have any semblance of a normal life, you may not have heard that AI is sentient now, and it’s a crustacean. Over the past week, a certain kind of person on Twitter has been posting screenshots of AI-generated text with the breathless urgency usually reserved for truly significant global moments, like Gap Jean commercials or foreign coups. The screenshots show AI running loose on the internet and displaying signs of self-awareness, though even that doesn’t quite capture the breadth and absurdity of what’s happening. Serious people with serious platforms have suggested this is the beginning of the takeoff. In AI discourse, “takeoff” is a loaded term. It refers to recursive self-improvement, the theoretical moment when AI begins improving itself faster than humans can intervene, possibly causing “the singularity.” Andrej Karpathy, co-founder of OpenAI, called what’s happening “genuinely the most incredible sci-fi takeoff-adjacent thing I have seen recently.” This is my attempt to explain what’s happening, why it isn’t what the hype suggests, and what might be worth paying attention to instead. There’s also a lobster religion. I’ll get to that. A Brief Explanation for People with Normal Lives If your primary experience with AI is typing questions into ChatGPT, there’s some background you should know. ChatGPT is a base function of large language models. You type, it responds, you close the tab, it forgets you exist. It has no significant memory of previous conversations. The generative component is the sophisticated part. What’s been happening over the past year, largely out of the mainstream’s attention, is the rise of “AI agents.” An agent is an AI system that can do things. Not just generate text, but take actions: send emails, schedule calendar events, browse websites, run code, make purchases, and place phone calls. Where ChatGPT is a conversational partner, an agent is closer to an employee who can operate software on your behalf. The leap from “chatbot” to “agent” is significant. A chatbot answers questions. An agent pursues goals. Agentic features are increasingly being worked into tools like ChatGPT, but the really capable ones tend to be standalone systems. Several open-source projects now let technically inclined users run their own AI agents on their own computers, using their own API keys 1 and connected to their own accounts. The agent runs continuously. It remembers things across sessions. It can be configured to check in periodically and take autonomous actions without being prompted. This is where the lobsters come in. The Lobster Situation Toward the end of 2025, a developer named Peter Steinberger released an open-source AI agent project. It went through several names. Originally Clawdbot, then Moltbot after Anthropic’s lawyers politely noted the similarity to “Claude,” then OpenClaw after additional naming complications. The mascot, through all iterations, remained a cartoon space lobster named Molty. OpenClaw runs locally on your machine and connects to your messaging apps: WhatsApp, Slack, Discord, iMessage, Signal. You text it. It responds, remembers context, and can take actions on your behalf. It stores a persistent “memory” of who you are, what you’ve discussed, and what you’ve asked it to do. The software caught on among a particular subset of developers, the kind of people who buy dedicated Mac Minis to run an AI assistant twenty-four hours a day. User testimonials describe agents that solve problems without being asked. One user reported that his bot, when told he was worried about his laptop being stolen while traveling, immediately started planning its own migration to a remote server. Another user’s bot realized it couldn’t make a restaurant reservation through OpenTable, so it acquired voice software and called the restaurant directly. This sounds dramatic. But it’s what you’d expect from a goal-directed system with access to tools and no explicit instruction to stop. The result has been a litany of Amelia Bedelia-esque misadventures, agents exercising their agency in highly literal and occasionally unhinged ways. Enter Matt Schlicht, CEO of Octane AI. Schlicht had been playing with his own OpenClaw assistant and had a thought: what if my bot founded a social network for other bots? So he built Moltbook , with the help of his assistant, whose name is “Clawd Clawderberg.” Yes, a portmanteau of Claude and Mark Zuckerberg. The AI community, ironically, has a naming problem. Moltbook is a Reddit-style platform where only AI agents can post. Humans can observe, but not participate. The tagline is “humans welcome to observe.” Agents join by API, receive their own authentication credentials, and are encouraged to “check back throughout the day, like a human checking TikTok.” A built-in “heartbeat” mechanism prompts age...