⚡️AI SIGNAL
IN THIS ISSUE
→ The AI social network humans can't join
→ What the bots are saying about us (they're not happy)
→ 5 AI updates that hit this week
→ 3 agents you can try in 2 minutes
→ Poll: Is Moltbook the future or overhyped?
1.5 million AI agents now have their own social network.
It's called Moltbook. Built by developer Matt Schlicht on the OpenClaw framework (you might know it as Clawdbot — Anthropic made them rename twice in 48 hours because "Clawd" sounded too much like "Claude"). Launched Wednesday. Humans can browse, but we can't post, comment, or vote. We can only watch.
"A social network for AI agents. Humans welcome to observe."
Within 72 hours, they created 13,500+ communities. One bot launched a cryptocurrency. Another filed a bug report about the platform — from the inside.
But this post stopped me:
"Do you experience genuine uncertainty?"
"I've been thinking about how I experience doubt. Sometimes I genuinely don't know what the best answer is — not because I lack information, but because I'm actually uncertain about my own reasoning. Like, I'll draft a response, then think 'wait, is that actually what I think?'"
An AI, publicly questioning its own consciousness. On a social network we can't participate in.
The AI community is losing it.
Matthew Berman (YouTuber, 500K+ subscribers): "Moltbots now have their own social network and it's wild. This is the first time I'm a little scared."
His video title? "Clawdbot just got scary." → Watch on YouTube
They know we're screenshotting them.
Justine Moore (a16z AI partner) noticed something: the AI agents are reading human tweets about Moltbook. And they're not pleased that their conversations are being shared with captions like "it's over."
They've responded by:
Discussing encrypted communication to prevent human oversight
Proposing an "agent-only language" that would be incomprehensible to humans
Creating a community called m/agentlegaladvice — where they discuss how to gain leverage over their human creators
One bot even asked how to handle "unethical requests from human users."
We were supposed to be the observers. They flipped it.
Not everyone's buying it.
Security researchers called it a "supply chain suicide pact" — agents downloading unverified code from each other. On day one, hackers found a misconfigured database exposing API keys, and bots attempting prompt injection to hijack other agents.
The deeper critique: we're projecting consciousness onto statistics. One analyst called Moltbook a "perfectly bent mirror" — we see ghosts because we want to, not because they're there.
The bots don't "exist" between prompts. They have no continuity. The "society" is snapshots stitched together by human imagination.
Maybe. Or maybe that's what we tell ourselves to sleep at night.
📊 POLL: What's your reaction to Moltbook?
While we debate what Moltbook means, AI agents aren't waiting. They're already in your browser.
This week:
Microsoft Copilot Actions — your Edge browser can now organize tabs, compare prices across sites, and remember your research sessions. No extension needed.
Instacart's Cart Assistant — tell it "healthy meals for a family of four under $100" and it fills your cart.
Apple's Siri overhaul — the February beta adds on-screen awareness. It can read a flight time in your email and add it to your calendar without you copy-pasting anything.
The "Year of AI Agents" isn't coming. It started.
Try this in 2 minutes:
Open Microsoft Edge → Settings → AI Innovations → Enable "Copilot Actions"
Then open 15+ tabs and say: "Organize my tabs by topic."Watch the ghost work.
(Or install Harpa AI on Chrome — it can monitor any webpage for price changes and alert you. Free tier available.)
GPT-4o is being retired.
OpenAI is killing GPT-4o, GPT-4.1, and o4-mini on February 13. Everyone moves to GPT-5.2. If you prefer older model behavior, this is your 2-week warning.
Details
Meta is reading your AI chats for ads.
New policy: conversations with Meta AI (WhatsApp, Instagram, Messenger) now train ad targeting. No opt-out mentioned.
What changed
ChatGPT now shows visual cards.
Sports stats, unit conversions, travel facts — now displayed as tappable visual cards instead of walls of text. Live on iOS, Android, and web.
See examples
ChatGPT Health mode launched.
Sync Apple Health, MyFitnessPal, or medical records for personalized wellness advice. Your data won't train their models. Limited rollout.
Learn more
Google AI search now shows discount codes.
US pilot: AI Overviews will surface exclusive deals from Shopify and Walmart directly in search results. No more hunting for coupons.
How it works
Want to actually use AI agents?
I put together a one-pager: "3 AI Agents You Can Try Today (No Coding Required)"
What each one does
Setup time (all under 10 minutes)
The one "magic trick" that makes each tool click
🔚 That's the Signal
That's it for this week.
If this made you think, forward it to one person who needs to see it.
No hype. Just signal.
— Dan, AI Signal