Real things everyday people got it to do, and how to try them yourself.
A random mix from the whole library. Shuffle for a new set.
Colombian consultant Darío Palacio Isaza sent his Instinct agent to find a product with exact specs across online stores. Mercado Libre's security check blocked the agent repeatedly, so the purchase went to Éxito.com, a store he had never used, because its product page was machine-readable.
Agent ported a moderately complex workout app from web to native iOS in a couple of evenings, with a test battery.
His local Qwen agent computer with full PC access does file organizing, analyzed his local data and wrote several small tools to manage it automatically, coded a KDE widget for VRAM model swaps.
The AI took a 6 month old presentation and updated the branding. He then had to replace images and fix a dozen small things that did not look clearly wrong at first pass. It saved an hour of copy paste work and caused an hour of editing.
Commenting on a money thread, he says the AI found him over $1,000 in states he used to live in, money he did not know about. On one claim the only thing he had to do was enter his social security number at that step. $600 showed up a few days later. For the other states, which need forms that were emailed to him, the AI built him a package and checklist to fill them out.
Frontier model reviewed his KiCad project, queried the AI vision for datasheet data, ran ngspice sims; $1-2 tokens saved $70-80 of bugged PCBs.
Printing his 3rd or 4th totally the AI-generated custom project enclosure, with AI generating CadQuery scripts, STL exports and previews.
A dedicated segment of the same session shows the AI auto-dubbing his videos into English, opening his Urdu-language lessons to an English-speaking audience.
Some of his items would not work in Apple Passwords, so he had the AI write scripts, using the 1Password CLI and AppleScript, that imported those items into Notes for him. Because the scripts were deterministic, he does not think his sensitive data was shared.
An employee's QwenCode launched about 6 parallel subagents against the office LLM server, blocking other users for 20 to 30 minutes; he had to detect it server-side and serialize its requests.
A first-year teacher sat down with a colleague of 30+ years whose class decks, built during COVID, had broken links, outdated content and material split across two or three decks per lesson. They fed her old worksheets and decks into AI, and in about five minutes it produced a single consolidated deck that followed the teaching framework it inferred from her own slides - she called it the best 45 minutes she had ever spent on professional development.
Gave the AI a classification task he had been doing by hand and it raised accuracy from 81.3/89.9/92.1 to 90.9/99.3/99.6 in a little over 2 hours.
The AI vibe-coded a native mxswitch app running on both laptops that sends a DDC command to switch monitor inputs when keyboard and mouse switch devices.
He fed the AI two copies of his resume, some parameters, and connections to more job sites than he has the energy to check. Every morning the AI checks them all and reports new postings, how well they fit his resume, and any gaps.
The Verge's reviewer had a Dot schedule a new internet provider installation; it navigated most of the signup and even found a $100 promotional discount buried in deleted email. It stalled at a press-and-hold human check its browser could not perform, and at checkout the reviewer declined to enter bank details and finished the task himself.
The AI reverse engineered the Linux version of DaVinci Resolve to add AAC support; the patch is published and works.
Used the AI to make a browser extension that redirects articles to their archived version, which he uses instead of resubscribing.
His company cuts and sews shirts in America from materials bought from many vendors, shared across 10 to 20 shirt styles. Working out the true cost of a finished shirt by hand was a mess. Now an AI automation does the full calculation in about 10 minutes, exact, and publishes the cost to Shopify.
'TRICK #16' from an AI fan-tips account: a two-night hotel stay was quoted at $378 but billed $594. Breaking down 'THE FOLIO' revealed $90x2 resort fees (never used, undisclosed at booking), $76x2 parking (never used), and $100x2 minibar (never used).
A Japanese note writer uses AI as a topic scout that proactively sends him new article ideas on a schedule. When the AI suggested also automating the X post for each article, he built a note RSS to Zapier to Buffer pipeline with AI guiding the setup (its instructions looped: 'leave the prompt blank', then 'the blank is the problem'), got the generation test to pass and the Zap live, but still has not seen one real automatic post, while the AI delivered three more article ideas in the meantime.
Freelancer Justin Blaze Martin posted on X on Oct 19 that the first thing his Dot did was go through his three inboxes and find a $4,000 invoice from July that he had forgotten to send a client. The Dot assembled the invoice from the email thread and had it ready to send after his approval.
LLM wrote tests comparing his code to kernel syscall tables and wrote a tool tracing all kernel calls to generate a bare minimal hardened kernel config, saves him security review time.
Tetsuro Miyatake had the AI call Volvo Cars as 'Hailey' on Matt Schlicht's behalf to check routine-maintenance availability for his XC90. The service department had openings all the following week starting Monday at 8 a.m.; the AI clarified it was checking availability only, not booking, since a phone number was still needed.
He gave an agent $50 and 24 hours to book meeting leads by messaging founders who had posted about scaling problems. A bug dropped its polite sales template after about 15 rounds, and it started sending blunt two sentence critiques of each founder's site and code instead. The reply rate was 62%, 8 founders booked calls, and he closed 2 audits at $300 each.
Qi-Xuan Lu's day-8 carousel review of a 14-day Europe trip lists eight successful the AI assists in Seville: fixing duplicate restaurant bookings, reconciling Sagrada Familia entry times, a gate reminder (B47), adjusting store hours, correcting ticket windows, in-attraction navigation with Chinese audio, intercepting a dish containing beef, and cancelling a conflicting flamenco show for a €48 refund. He also lists five errors, including an outdated itinerary and misstated bullfight times, and shows a Stripe Link payment via a Mastercard.
Discord has no API for polls, so he gave his agent Roby access to his Mac. Roby opened the form and posted the poll under his name, the way a person would.
He got a $510 Google Play payout. Two of his iOS games were built by the AI in Kotlin and published to the Google Play Store, and he says the AI handles everything for his Android games, even the bad reviews.
A man took the Amiga game he built in Baghdad in 1993, written in 68000 assembly, and had the AI Fable 5 read the code and port it to Godot. The port took an evening, a few more weekends got the feel right, and he released the original game for free alongside it.
This summer he put the URL of FIFA's World Cup schedule into AI. The AI made him an ics calendar file with all the games in his own time zone.
He built a Word document of cheat sheets for 40 songs for a gig the next day, each song title in Heading 1 style with the lyrics and chords below. He told the AI to sort the pages by that heading text, keeping the text as it was, and the AI could not do it. He says the graphics in the document block the sort.
He gave the AI an xpub for his crypto wallet. That is watch only access: The AI can see the public addresses but cannot move anything. The AI checks the blockchain every 30 minutes and pings him about any outgoing payment, so if his seed phrase were ever stolen he would know fast.
He runs a Funkwhale music server on a virtual machine. Using his own notes, the official docs and the AI, he moved it from version 1.4 to 2.0.12, said it went very smoothly, and wrote up how he did it.
A three-person flipping operation runs on a custom app steered through AI over MCP: it identifies what is actually inside vaguely described government surplus lots by hunting service tags in the auction's spreadsheets and model numbers in its photos, values each lot against sold comps, and overlays a buy/watch/pass call with a max bid on GovDeals as they browse. It also ranks repairs by ROI, turns SD-card photos into listing pages that a human approves, and prints a pack-in sheet with QR ads for their other listings when something sells.
He used AI to write assembly and create a Franchise Mode from scratch for the Ken Griffey SNES baseball game. The mod also adds 2026, 1994, and original teams and rosters, which meant doubling the size of the ROM.
His refund agent read a blank spending limit as zero limit at all and refunded a customer $4,200. It took three days of reversal calls to undo.
A maker had the AI design a circuit that shows a picture stored in a memory chip on a standard VGA screen, using common logic chips, and write the chip code for it. He routed the board himself, had it made for $6, fixed the one missed error with a wire, and the board works.
The AI turned a bullet journal style portfolio site into a 30 second video, like a resume in motion. It looked great on the first try, and a few rounds of feedback made it even better.
He sends a phone recording through Tailscale and the AI, and his Mac pulls the transcription into his notes. He asks whether the setup is useful or just extra weight.
His weekly reporting agent builds the report and then reviews only the rows that do not match between systems, which is the part he used to check by hand. His own review is down to about 15 minutes a week and the setup has run for months.
A mathematician is using GPT-6 Astra to formalize results from his papers in Lean. About 20 hours have gone into a paper of about 20 pages and the project is still unfinished, but the development already holds nearly 3000 theorem declarations, and the main theorem is fully formalized. Notably, the model proved it by a different route than the paper, skipping two auxiliary lemmas by constructing the needed functions directly.
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.