A random mix from the whole library. Shuffle for a new set.
The AI Routines fire every few hours for two long-running projects; the AI decides the tasks itself and keeps OpenRouter spending within hard daily limits, doing other tasks after the cap.
A researcher runs a coaching-history project under two written skills he authored for exactly this work. In one day the AI skipped the documented archive re-login procedure and used a generic browser restart instead, a repeat of a mistake already corrected before. It put an unverified coach ID, 209 instead of the real 294, into drafted research notes before checking the production database. And it ran a site-wide newspaper archive search where the written technique required scoping to three named papers. Every error was against a procedure that already existed in writing.
A creator gave the AI a screen recording and asked for a setup tutorial; the agent produced a HeyGen video presented by her own avatar. It kept her preferences in memory for follow-up video work.
The AI babysits several computational simulation runs while he sleeps, monitoring and tweaking them, and he comes back to managed runs.
His quit vaping app Puff Counter passed 10,000 downloads with a 4.1 star rating after he built it in 2024 and never advertised it. Last week the AI rebuilt the app's whole website in one day.
A Korean automation instructor sells an AI skill kit for Naver Blog and documents his own test on a real shop: he handed the skill five shop photos and a one-line memo, and it wrote the post, placed photos, a video, a map and tags, and completed a scheduled publication in 3 minutes 20 seconds. With a reservation set, Naver itself posts at the chosen time even with the computer off.
Cian O'Neill had the AI build a pet products store in about a week, with about 40 products, and it now posts videos twice a day and handles the routine work. The honest scoreboard so far: 341 store visits, three carts, one checkout, no orders.
Boston demolition startup Circa built Waste-Less Vision with AI: customers submit photos or video of a space, the system identifies materials and prepares a preliminary estimate of disposal costs and labor hours for human review. A 15-minute basement video that once had to be watched fully now becomes a line item automatically.
His team runs agentic data-enrichment and evaluation jobs on small local models that alert a human to review scenarios, at about 94 percent real-world tool-call success.
His company runs three coordinated agents on its product account: one watches mentions and a watchlist, one writes posts and replies, and one controls quality. Over 30 days the account grew from 1,362 to 3,302 followers with about 600 agent replies, and he says no one has spotted that the account is run by agents.
Every idea is a real post from someone who tried it, summarized in our words, with a link to the original. App names are swapped for plain words.