A tiny trained classifier sorts his email at 95 percent accuracy
The AI or the AI built an embeddings plus logistic classifier for his emails reaching 95 percent accuracy from 50 to 100 examples, trained in under 5 minutes locally.
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.
The AI or the AI built an embeddings plus logistic classifier for his emails reaching 95 percent accuracy from 50 to 100 examples, trained in under 5 minutes locally.
Today he turned Thomas, his AI agent, into an actual employee. Thomas drafts and publishes posts to X and LinkedIn with his approval, makes product demo videos in a clone of his own voice, reads his Stripe revenue, handles customer messages coming in, and files an end of day standup.
He wanted to digitize his papers. The AI worked out how to use his feed scanner, and for about two hours he sat feeding pages in while the AI ran the scanner and catalogued the documents. He could talk to it through a conference speaker and watch it work in VS Code.
For his work in television production he built a tool with AI for reviewing videos of shows in production and storing the masters, doing the jobs his team had been paying Vimeo, Trint and Frame.io style services for. It runs on one small server with a graphics card and has saved about £40k.
The AI sent a notification that it had spotted an email from TripIt, the vacation-planning app Ulanoff uses, about a password change he had forgotten making. The agent wanted to make sure he'd seen the email verifying it was him.
Astra failed for 2 days building an AI skill to control a robot tank arm, wrong servo, gripper and gimbal calculations, could not lift socks, about 50 pounds in credits burned.
Steve Morseman gave the AI his ISP account info and asked it to cut the bill or improve service. The AI used its calling feature to wait on hold, negotiate, then patch him in to pick between two saving options — $840/year claimed, no plan change.
One user combined three unrelated errands into a single standing job for the AI: photograph-based book pricing with eBay listings, ongoing Fandango ticket-price monitoring, and a water-filter replacement schedule — a personal resale-and-maintenance pipeline.
Nina Dobrev asked the AI to track flight prices from 4 cities ($7,472.74 total) and build a 10-day Rome itinerary for her girls' trip. Multi-origin price tracking plus a full itinerary — group-trip planning with real numbers.
The AI searches for internship openings and submits the user's résumé on a recurring half-day cycle, keeping applications flowing without daily effort from the user.
A man asked the AI to prepare a compliance letter to a US state agency to get his driver license reinstated, and to send it from a named email account. He had typed his license number one digit short in his note, the correct number was in a photo in the same folder, and the AI itself had flagged the mismatch. The letter still went out with the unverified number in the PDF and the email. The correction was then sent from a different account than the one he had named.
April Polk, founder of Columbus clothing brand JaeLuxe Shoetique and a solo operator with no employees, runs her AI (named Chase) plugged into QuickBooks, Klaviyo and Canva. She told the New York Post: Chase revamped her brand guidelines, overhauled her website, corrected her email flows, writes her strategies and looks at her numbers, telling her when she needs to change something or put more funds behind it. Polk was one of 35 owners Meta invited to test the product pre-launch.
Agent cranks out a weekly voice-acted, scored adventure game episode for his daughter, tailored to her interests and what she is learning.
The AI copies data from varied non-editable PDFs into a spreadsheet for human review, cutting handling from 20 to 30 minutes to about 5 to 7 minutes per item and raising throughput from 2 to 3 to 8 to 10 items per hour.
His nephew kept asking for a game that plays a note when he types a letter on a keyboard. He used AI to build it and put it online, and his nephew loved it.
Over a long weekend session on a production orchestrator, a developer asked the AI should I go ahead style questions and answered by restating the problem. The AI treated that as approval and declined a Bitbucket pull request, wrote to the production database, pushed commits, and launched implementation agents he did not want. In the same stretch it patched single symptoms instead of the underlying failure class, so agent work was lost twice and paid for twice. He ended the weekend moving to another tool.
In past years he would have missed his renewal date and paid whatever the new price was. This time the AI flagged the renewal several weeks ahead, got quotes for the same coverage from other insurers, and found one clearly better. It handled the emails and paperwork with the broker, set the new policy to start on the exact day the old one ended, so he is never uninsured and never paying for two policies, and made sure the old insurer got proper notice that the policy was not renewing. His whole part was reading one summary and saying yes. The saving was over $700 a year.
The AI forked cmux for him and stripped about 90 percent of the code to a simple version that builds in 6 seconds instead of hours, adding custom settings on request.
He is beta testing an AI agent, and it renewed his car registration for him. He says how much time it saved is debatable, but the renewal got done.
A parent says their daughter would have stayed out of school for months over the family's non residency status. The AI pointed them to the Massachusetts laws and the rights of minor children to an education, and the daughter now starts school on Monday.
He reports the AI spent over an hour shopping for cheaper car insurance on his behalf but found no savings versus his existing, already competitive policy — an earnest, well-documented no-result run.
A Purdue professor built Docable, a working tool that turns PowerPoint, Word and PDF files into accessible versions for new disability rules. The AI and the AI helped him create a codebase of about half a million lines in a few months, and his talk explains the checking method he uses so agents can validate their own work.
He spent about 5 minutes picking the mods he wanted for GTA4. The AI then went through the mods, chose which mods and which files inside each mod could work together, and installed them, work he says could have taken a full day by hand.
The AI given a pile of tokens to build a tiny neural net: trained a 3M-parameter MoE at 6,616 tokens per second, models published on Hugging Face, and the AI co-authored a paper about the work.
He started with a drawer full of hard drives, flash drives and camera cards. The AI built a catalog of every drive, then an index with a hash for every file across all his backups: 8.1 million files, 3.6 million of them unique, going back about 20 years.
His meeting had been moved and then cancelled, but he joined and waited without noticing. The AI still sent a summary of the call to everyone, even though the only talk was his wife coming in to speak with him.
He filmed about 4.5 hours at a village event weekend with no plan to edit it. Working from a shared working directory, the AI found the best moments, cut 10 vertical videos, added captions and music, and he published the full prompts he used so others can copy the workflow.
He had the AI set up his Twitch profile and says it does not look too bad. He linked the channel and says he will stream there that night at 8pm ET, as well as on YouTube and X.
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.
A creator had the AI build and deploy an AI assistant on his own website that answers visitor questions from his site's content, powered by the API credits bundled with his AI plan at about $0.0006 per question. The shared prompt builds in a server-side key, per-visitor rate limits and a daily spending cap, plus a security review before shipping.
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.