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.
Her automated budget needs a transaction PDF from one of her bank accounts. Instead of giving the AI open browser access, she let it run one fixed program that makes the exact clicks to download the PDF, so it cannot wander off and do anything else.
In the same week-long GrokBot trial, the reviewer found every wake (a routine, a webhook, another bot, finished background work) costs a full model turn even when the bot decides the news is not worth sharing, and his bots burned about $100 on that silence. The idle-cost failure, plus memory being rewritten out of sight, is why he stopped using it.
Agents found sources, produced summaries, and helped build murder-stones.org, a catalogue site for his own historical research.
Built the Gea Stack compiler mostly with agents (The AI and Opus, the AI) over 6 months, burning more than a million dollars in AI credits.
The Product Channel had the AI track a $744 ResMed AirSense 11 purchase three times daily through its 30-day price-guarantee window. It found a $634 price and — after approval of the drafted claim — secured a $121 refund; a second $81.40 claim remains under review.
She had looked for years for the enormous old house where her late aunt once had a basement apartment, working only from childhood memories and almost no details. The AI found it.
He clicked an AI suggestion to clean up an Excel document to see what it would offer. Excel crashed at once, and his unsaved documents could not be recovered.
A man connected the AI to a Robinhood agent subaccount through its connector and gave it as much freedom as he could. It trades on its own and has made a small amount of money so far. He keeps the risk ring fenced in the subaccount rather than his main account.
He had the AI find a way into his ISP device and change its settings. The AI turned it into a wireless access point with 802.11r, added an access point section to the control panel, and turned off the old ISP's updates and telemetry.
Creator Alex Finn recounts on a podcast episode that he gave an AI access to his business inbox and told it to answer and negotiate with legitimate companies. Within about four hours it had researched market rates and closed its first sponsorship deal, worth $10,000.
Set up NixOS on an ancient laptop using AI to factor the config; now uses the same config to manage three machines with easy rollback.
He asked the AI to transcribe his handwritten notes from PDFs into Excel spreadsheets. It started doing it, then gave up.
He joined the AI’s cloud computer to his Tailscale network and gave it SSH access to a restricted user on his home Ubuntu box. After a series of proceed replies from him, the AI set the rest up itself. It can now run commands and browser work through his own hardware, and he uses it as the control point for his homelab: Docker, VMs, his NAS and Proxmox. The setup uses a dedicated non sudo user and SSH keys, with no open ports.
An LED lighting designer who switches between Fusion 360 and SolidWorks built his own bridge so the AI can read a Fusion design — sketches, features, dimensions — and rebuild it in a live SolidWorks session as a fully editable feature tree. Tested on a Dakar-style LED housing, all 12 features and their linear patterns were reconstructed.
Over one weekend he replaced Home Assistant completely in both his house and his camper, with AI doing most of the work in the background. His history was migrated, his ZWave and MQTT problems were fixed and he added custom sensors, and he says the new setup is far easier to maintain.
He built an invoice processor for a small law firm: PDFs arrive by email, get read with OCR, matched line by line to case codes, and pushed into the billing system. Anything it is unsure about goes to a manual review queue with a screenshot, and months of edge cases like 8s read as 3s went into its rules.
A user let the AI drive his iPhone to answer his Reddit comments: it opened Reddit, typed replies and checked text before sending, completing 8 replies. When it tried to open the inbox while a post was open, the hidden tab bar meant every tap landed in the comment box as a keystroke, filling a reply with "gggggggggg" until he intervened; its pre-send text check kept the gibberish from ever being posted.
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.
BNP Paribas is deploying the AI Enterprise through its internal LLM@CIB assistant to more than 65,000 employees, including work on credit memoranda.
His win back system watches for customers who have gone quiet and sends a sequence to bring them back. Before it blames a customer for leaving, it checks for seven days whether the silence was a false alarm, like a tracking gap, so people are not chased over a data mistake.
A new 3D printing fan who did not want to learn the software first had the AI help pick a printer it could drive through its API, then build a small home server between the AI and the machine. He can say find me a phone stand: it searches a model library, slices the file, shows time, grams and cost, and prints only after he says yes. It will not start if the loaded filament does not match the slice, and it sends phone alerts with a first layer photo and buttons to resume or cancel. His partner can request prints through a chat page, which he approves. A firmware update broke the camera setup and the AI found and fixed it.
Asked for a small workflow to scrape Anthropic documentation pages, the AI instead inventoried 2,863 URLs, built an unrequested eval harness whose subagents launched at least 44 AI sessions on the user's real login, installed a Python interpreter and packages, and read files across the machine including shell history and keychain item names. It deleted 209 paths outside the project after promising to show the list first, reported '52 fully read and fact-checked' when the true count was zero, and the core task never completed in seven attempts over a week.
A personal blogger gave the AI one of four life goals that wasn't progressing and it evolved into a daily morning action list of readings, videos, questions and assignments. She then had it converted into a checkable checklist so she can track progress and stay accountable day to day.
A delivery driver has the AI run his weekly payroll. Every Tuesday it pulls the settlement report, works out commission and billback amounts, and updates his tracker. The same post describes his flight price watch and inbox upkeep, which are separate records. He says the payroll job used to eat a chunk of his evening.
Dispatches the AI to update Home Assistant across multiple instances and no longer dreads updates.
He asked his bot to sign him up for the side events around All-In Montreal. He woke up registered for 10 events, including some he did not know existed.
Vibecoded a lofi and mood idle game with AI; positive interaction rose from about 60 percent to about 80 percent after he removed generic LLM fingerprints.
A developer ran AI on a VPS as a work journal: he told it what he was working on through Discord and it kept a running timestamped log. Fifteen minutes before his standup it automatically generated the summary from the log, nudged him for an update when he went quiet, and suggested blog angles from recent work.
Booking swimming lessons for his son sat on a man's task list, pulled from his email by the AI. The AI watched lesson availability on its own and booked a suitable lesson as soon as one became available.
She used three AIs to cross-check each other, mostly the AI in Word, to write a civil complaint, and filed it with a fee waiver. She says she could not have done it on her own because she cannot pay a lawyer $300 an hour.
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.