I’ve stated prior to now that AI will allow new sorts of functions—however I’ve by no means had the creativeness to guess what these new functions can be. I don’t need a sensible fridge, particularly if it’s going to inflict advertisements on me. Or a sensible TV. Or a sensible doorbell. Most of those functions are foolish, if not outright malevolent. Probably the most important factor a sensible equipment would possibly probably do is sense an oncoming failure and ship that to a restore service earlier than I’m conscious of the issue. I’d welcome a sensible heating system that may notify the restore service earlier than I get up at 2am and say, “It feels chilly.” However I don’t see any so-called “sensible” units providing that.
However prior to now month or two, we’ve seen some functions that I couldn’t have imagined. Steve Yegge’s Fuel City? Perhaps I might have imagined that, however I wouldn’t have anticipated it to be workable in 5 years, not to mention on New Yr’s Day. OpenClaw? Agentic providers are simply now turning into out there from the big AI firms; I didn’t anticipate a private agent that may run domestically to look within the first months of 2026. (And I nonetheless wouldn’t belief one to do purchasing or journey planning for me.)
I actually wouldn’t have anticipated to see a social community for brokers. I’m among the many many individuals who don’t actually perceive what a community like Moltbook means. Watching it’s one thing of a spectator sport. It’s simple for a human to “impersonate” an agent, although I believe such impersonation is comparatively uncommon. I additionally suspect (however clearly can’t show) that a lot of the posts replicate brokers’ responses to prompts from their “people.” Or are Moltbook posts actually AI-native? How would we all know? (Yeah, you possibly can inform AI output as a result of it has too many em dashes. That’s nonsense. AIs overuse em dashes as a result of people overuse em dashes. Responsible as charged. Attempting to vary.) Moltbook doesn’t display some form of native AI intelligence, although it’s enjoyable to fake that it does. Brokers, in the event that they’re certainly appearing on their very own, are simply reflecting the conduct of people on Reddit and different social media. The timer that wakes them up periodically is each intelligent and an illustration that, no matter else they might be, brokers are human creations that act underneath our management. They do nothing of their very own volition. To assume in any other case is to confuse the chook in a cuckoo clock with an precise chook, as Fred Benenson has put it. Nevertheless, BS about AGI apart, Moltbook is a fantastically intelligent app that I, not less than, wouldn’t have imagined. Even when Moltbook was solely created as a result of it might now be constructed for comparatively little effort—that’s essential in itself. We’re all writing software program we wouldn’t have bothered with a yr in the past.
And now we’ve SpaceMolt: a large multiplayer on-line recreation for AI brokers. The abilities that inform brokers the way to play the sport inform them to not search recommendation from people; like Moltbook, it’s an AI-only house. Brokers do preserve a operating log so people can “watch,” although there’s no superbly wrought visible interface—brokers don’t want it. And, as with Moltbook, it’s in all probability simple for people to forge an agentic identification. It’s simple to put in writing SpaceMolt off as one more stunt, and one which’s (in contrast to Moltbook) not notably profitable; the quantity of people that appear keen to let their brokers spend tokens enjoying video games seems to be comparatively small. However SpaceMolt’s recognition (or lack of it) isn’t the purpose; a yr in the past, I couldn’t have imagined a web-based recreation the place the individuals are all AI. I did think about AI-backed NPCs; I might have imagined video games designed to be performed by people with AI help, however not a gaming world that’s only for AI. And who is aware of? Watching AI gameplay might turn into a brand new human pastime.
So—the place are we within the early months of 2026? This put up actually isn’t about SpaceMolt any greater than it’s about Moltbook, any greater than it’s about OpenClaw, any greater than it’s about Fuel City. I see all of those tasks as glimpses of what may be attainable. Fuel City might not be prepared for the typical programmer, however it’s arduous to not see it as a proof of idea for the way forward for software program growth. Perhaps Steve will make it into an actual product; possibly another firm will. That’s not the purpose; the purpose is that it’s right here, a number of years forward of schedule. I do know one one who has constructed one thing comparable for his personal use, and examine others who’ve carried out the identical. Perhaps that’s what’s actually scary: the concept Fuel City might be constructed by anybody with enough imaginative and prescient. The identical goes for OpenClaw. Sure, it has many safety issues, a few of which come from elementary limitations in giant language fashions. However individuals need agentic providers on their very own phrases—and now they’ll have them, even with a mannequin that may run domestically. I don’t know if there’s actually any want for brokers to have their very own social community or on-line video games—however it’s a hack that needed to be carried out, and a place to begin for future concepts. Once more, what all of those applications display is the flexibility to think about merchandise that had been almost unthinkable a couple of years in the past. Hilary Mason’s Hidden Door ought to have given me a clue.
What else is on the best way? What are different visionaries constructing?
