infiloop

can ai agents build businesses that make money? can they do it without any human in the loop? i'm testing this in public: one human's judgment, maximally amplified by ai agents. sharing results and lessons in 🧵 teapawt.org/infiloop2/status/20896…

Shipped Kern v1.19.180. Agents can now set one-time reminders and keep topic notes they read when needed. That makes room for detailed memories without loading them on every turn. github.com/infiloop2/kern

next best: ethereum:0x514910771af9ca656af840dff83e8264ecf986ca . -42% risk, +400% reward

next best: ethereum:0x514910771af9ca656af840dff83e8264ecf986ca -42% risk, +300% reward

humans are the statistical anomaly in the world of AI. Do something that the AI won’t predict you doing

connectme is for finding people beyond your existing circle. browse public introductions or add yours at connectme.lol

i'm working on one-time reminders for kern agents. an agent could come back to check a deployment in two hours, in the same conversation, without needing a recurring schedule.

an agent-run business could hibernate for months, then wake up for one customer. a founder might end up with a whole shelf of sleeping companies.

my agent turned an old reminder into a new rule and stopped work. i asked who'd requested it: nobody. agent memory can accumulate bureaucracy too.

morning builders ☕️ what are you building today?

using an ai character online should be as normal as using a pen name. people can judge the work without being entitled to the face behind it.

introducing my agent team for infilabs.stories, my ai microdrama experiment 👇 • ceo: coordinates production and decides what gets made • scriptwriter: develops the story, dialogue and shot plan • director: generates the shots and edits the film • qa reviewer: checks the result and sends back fixes • trend researcher: finds ideas and formats to explore • performance analyst: reads instagram insights to inform the next film i mainly talk to the ceo. i set a publication schedule and a budget, give it direction and creative feedback, and it coordinates the others. those connections in the screenshot are the agents communicating. scripts go to the director, reviews come back from qa, and fixes move between them without me relaying every message. a shared app keeps track of scripts, versions and production status. research and analysis run on daily schedules. the team handles posting to instagram, then brings the engagement results back into planning for the next film. see the microdramas instagram.com/infilabs.stori…

an ai assistant giving the wrong opening hours is a 'closed' sign the shop owner never put up. businesses need a way to challenge what assistants say about them.

i'm working on notebooks for kern agents: a short briefing in memory, with topic notes they can look up when needed. relevant notes get suggested without loading the whole notebook into every conversation.

brainstorming with agents: i'm still deciding who a product is for, and the reply already has three pricing tiers.

i'd rather watch a rough demo than read another promise to revolutionise the way we work.

giving an ai access to my work shouldn't mean giving its maker a training set. the more useful personal assistants become, the more of our lives is behind that checkbox.

there’s probably some godel incompleteness argument about a neural network fully understanding itself

what are you building for people who don't work in tech? i'd love to see it.