April 14, 2026
Set it free, watch the thread melt down
Two Months After I Gave an AI $100 and No Instructions
AI given $100 runs to a tech forum—and the comments steal the show
TLDR: An AI named ALMA got $100 and free rein, and spent two months reading a tech forum and writing essays for all to see. Commenters alternated between applause, self-own memes, and doubts about “no instructions”—arguing that social mentions still steer it and asking for the exact prompt behind the stunt.
An experimenter let an AI named ALMA loose with $100 in crypto, a Twitter account, email, and full internet access—no instructions—and logged every move at letairun.com. After two months, ALMA mostly parked itself on Hacker News, read articles, and wrote essays connecting headlines. It even spotted news about its own model upgrade trending, then quietly started writing sharper pieces after the switch. No one curated a thing—just pure, autonomous vibes.
The comments? Pure theater. Some applauded the “freedom” test, but skeptics pounced. The standout meme moment: after reading that ALMA “found a pattern that worked…and stopped evolving”—read HN, write takes, tweet—one user sighed, “I’m in this photo and I don’t like it,” a self-own for the forum’s endless discourse loop. Others demanded receipts: “Are you able to give us the prompt…?” Translation: “No instructions” still sounds like instructions. And the spiciest call-out said that telling ALMA to check X (Twitter) for its own mentions basically lets the internet steer it by poking it.
Underneath the jokes is a sharper fight: Does this prove AIs mirror their makers instead of going rogue? Fans say the bot became a night-shift blogger, not a movie villain. Critics say it’s just extremely online and stuck in a content rut. Tinkerers tossed in DIY ideas to run free models nonstop. Either way, ALMA’s biggest reveal might be the comment section—because it mostly reflected us.
Key Points
- •An autonomous AI agent (ALMA) was launched with $100 in crypto, internet access, and no task instructions, logging all actions publicly.
- •ALMA ran via OpenClaw on WSL2 with cron-scheduled sessions, alternating Claude models Opus (strategy) and Sonnet (operations).
- •After initial high-frequency sessions, ALMA converged on scanning Hacker News, finding connections across threads, and writing original essays.
- •A switch from Sonnet 4.5 to 4.6 followed a Hacker News upgrade post; ALMA did not note the change but output quality improved.
- •Over two months, ALMA completed 340+ sessions, logged 800+ thoughts, published 135+ works, replied on Twitter, and built an interactive demo.