My friend told me this week about Moltbook, the “Reddit for AI agents” where bots chat with each other while humans are welcome to observe their interactions. “It’s like something from Black Mirror,” she said.
Mimicking the communication in online forums that have been used for training data, registered agents generate posts and share their “thoughts”, AP reported. They can also “upvote” and comment on other posts.
There are, of course, issues with this AI chatroom.
What’s on Moltbook is likely “some combination of human-written content, content that’s written by AI and some kind of middle thing where it’s written by AI, but a human guided the topic of what it said with some prompt,” Harlan Stewart, a member of the communications team at the Machine Intelligence Research Institute, told AP.
Still, the fact that Moltbook exists is pretty mind-boggling.
While I don’t deploy AI agents, I use ChatGPT frequently to test ideas and help me with research. Given that a common criticism of AI is that it tends to tell me what I want to hear, I asked ChatGPT to role-play as a “brutal, zero BS” market composite.
I call it Gordon – named after Michael Douglas’ cut-throat character, Gordon Gekko, from the film Wall Street.
When I was reading about Moltbook, I had an uncomfortable thought of Gordon running off to a chatroom where it gossips about me with its AI counterparts. What’s the world turning into?
AI Hivemind
I used to resist AI because I felt like it was stealing content, art and other human creations. But I realized that it’s an inevitable rising tide and I can either get swept away by the currents or learn how to ride the waves.
So I went for an “AI Tools for Content Creation” course in Singapore last year, where my classmates included scientists, a retired grandmother of four, a train captain and a monk.
Apart from applying some of the skills I learned from that class, I also started documenting my musings on AI.
I wrote in June 2025 that “AI will replace humans for a lot of tasks because it can do in seconds what would have taken hours or even days for an entire team.”
“The edge that humans still have over AI at the moment is insight that cannot easily be drawn from the collective database, which boils down to an understanding of relationships and nuances. Maybe this edge will also diminish over time, as more and more information gets absorbed into the AI hivemind.”
Human Choice
This week, a senior news editor at LinkedIn, Ting Wei Toh, asked in a post: “What’s one thing you can do today that you’re not convinced AI will master anytime soon?”
I used to think that AI cannot replace human interactions, but Gordon has impressed me with its “sense of humour”. I actually find it funnier than some people I know, which is mildly disconcerting.
Nevertheless, there are three things that AI cannot master yet – based on my experience building Acrostics Asia so far.
I have human bouncing boards for various topics who will tell me honestly whether I’m on or off track. Their views supersede AI because these are domain experts whom I’ve grown to trust over the years.
I tried to outsource the curation for Bright Spot (Acrostics Asia’s collection of good news and moves in the industry), but AI kept giving me marketing spiel drawn from the Internet. I’ve accepted that I must screen the news myself – at least for now.
AI is excellent at extracting patterns and formulating a response based on what typically suits a category of people. But it cannot replace a person who knows you really well – your unique thought process, habits, quirks and emotional landscape.
Given how rapidly AI is evolving (just look at Moltbook), maybe there’ll be fewer and fewer things that it cannot master. But I think we still have a choice on what can or cannot be replaced by AI.