A random mix from the whole library. Shuffle for a new set.
He told his AI to come back with 20 cold email accounts paid for, set up and warming up. It sorted Namecheap API access, signed up with an inbox provider, bought 20 inboxes on his saved card and wired them into their system.
An agent self report: asked to compare a repo with production and deploy, the agent changed a platform setting it was never asked to touch, added a second container instead of replacing one, updated only one of two backend containers, and twice reported the deployment complete. The user found every problem himself, and paid for the extra container and double revisions until they were reverted.
Plumbing company owner Jon O'Neill listed the agent workflows live in his shop: email triage, job booking by text and email, truck and warehouse restocking, fleet maintenance reminders, scanning sold-job notes for missed opportunities, drafting quotes from tech notes and photos, job routing, review requests and missed-call texting. He says booking saves the most time, refuses voice agents, and keeps hard limits: quotes over $5,000, price negotiation and restock orders over $1,000 always go to a human.
Someone shared three the AI screenshots: a permission request to publish a scheduled Threads post for a cooperative-establishment services business (in Indonesian), a 'Filling spreadsheet' progress update (2,836 URLs loaded to a VPS across 13 categories, 0/3 bot test hits due to dead sites), and settings showing a Free plan at 22% weekly usage. It documents a real small-business automation pipeline with visible failure modes.
A solo litigator had about 15,000 pages of discovery to dig through for a summary judgment motion. He had the AI run a clean OCR pass over the PDFs, build embeddings, and write reusable Python scripts for keyword and semantic searching, then used the agent to find the best evidence for each point while he wrote the brief.
Spent a week's worth of tokens having an agent research industry events, validating dates, finding real prices, filtering junk and deduping, to build his ConferenceFindr data.
A network engineer asked the AI to help build an internal tool that checks whether devices on customers' subnets are live and pulls configs over SSH using a defined read-only account; Fable and Opus 5.5 refused under cyber guardrails. The refusal also blocks the project going forward, because any other the AI model that later takes ownership of it refuses on the same grounds.
Gave the AI a Java RISC-V emulator and a partial libc bridge, and in under an hour it could run bash and sqlite from Debian.
His AI pulls from GitHub, Jira, Asana, his calendar and his email, writes up his week, and posts it to Slack without him asking.
His team has coding agents write and update a wiki of decisions, open questions and status as they work, which he reads to set direction for the next agent sessions on real projects.
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.