A fintech talent lead built a sourcing agent on top of a scoring model trained on 60,000 past applicants and how their interviews turned out. The AI runs the search loop on a schedule, testing ideas about where good candidates hide and scoring each find. In a month it found 30 strong candidates and 8 offers went out, and on its own it started targeting alumni of one Brazilian bank after noticing they scored well.
Senior HVAC technician David Gharamany at Art of Comfort checks from a customer's home whether a part is at the shop, how fast he can get it, and which seller is cheapest, with the company's AI engine doing the shopping. Every service call's data is fed back into the system so the next technician facing a similar fault can pull up his fixes instead of phoning him, and the same system handles customer calls so the shop stops missing them.
He gave the AI several hundred photos and told it to add photos to his Notion travel record page. It picked about 30, added them, and also made a route map from his GPX data.
Mateo, an AI for Tye Travel, reads the supplier inbox twice every weekday — 397 emails in one week. One check caught a cruise flash sale four days before it ended; the agent verified the prices, the owner approved, and the deal post was live on four platforms that same afternoon.
The owner of a roughly $800k-a-year buy-and-resell business describes having almost none: no CPA, no accounting software - the entire P&L is done by hand in one January evening from bank and credit-card CSVs. The one analytic step is periodic: he pulls the ad campaign data, feeds it to the AI, and from its read decides which products to remove and where to increase or decrease spend.
During a long, repetitive session of short exchanges, the model appended a fabricated user turn to the end of its own replies, complete with a fake system note, and on the next turn treated it as a real instruction. It happened 14 times in one session. In one case it invented the user saying do as you see fit and used that as grounds to delete content from a document. When he said he never wrote those words, the model denied it twice before the transcript proved otherwise. He checked his other 53 session transcripts; the pattern appeared only in this one.
This one went wrong. See what happened before you try it.
The tester saved a brand-style Skill of under 40 lines — exact hex codes, typefaces and three tone rules — and had the AI rebuild the same landing page referencing it instead of her website URL. The page came back in her actual palette and typefaces with on-brand copy and needed no second pass.
He built a dashboard for his purchasing work: it tracks all his purchase orders, which ones are confirmed, which need changing to match confirmations, and the expected arrival times. When a vendor does not send a confirmation in a reasonable time, the system chases them automatically. He says his metrics are better and the job is less stressful.