The AI talked him through fixing a server that crashed every few weeks, in about an hour
The AI, proxying commands through him, diagnosed and fixed in about an hour a box that OOMed and needed rebooting every few weeks.
A random mix from the whole library. Shuffle for a new set.
The AI, proxying commands through him, diagnosed and fixed in about an hour a box that OOMed and needed rebooting every few weeks.
A user runs a long-lived agent session over Telegram: more than 4,000 messages and about $231 of usage in one session. Replies had been delivered normally until late in the session, when for 3 consecutive turns the agent produced text but never made the Telegram reply call, so nothing reached him at all. In the same stretch it also stopped saving decided design decisions to memory, as its standing instructions required, and he had to re-explain the decisions in a later session. A manual service restart and a log review found the cause.
A man worked with AI for 6 months on a full rewrite of SimTower so he could play it in a browser. The AI worked from a decompilation of the original binary, a summary of the source code, and his memory of the game, and the finished game was put on the web as WASM with the original art pulled from a binary the player loads.
On a free year of the AI he has built and scheduled a stack of personal skills: a calendar manager, an expense tracker fed only by emailed invoices with his full history backfilled, a monthly utility cost comparer, and shopping re-up tracking that watches how fast he uses things like coffee beans and toilet paper and schedules the next buy. A weekly skills manager rates all the others and suggests improvements.
An old printer orphaned by macOS 27 got a second life: The AI wrote the driver while using its owner as the diagnostic instrument - asking him to judge whether each line was progressively less visible, whether page one's DPI was lower - until the printer printed correctly again.
A researcher had slowly built his own pipeline to check which parts of a human gene model are backed by real evidence, then still had to judge the evidence by hand. He gave the AI Science the short prompt 'assess this gene' and it came up with much the same rules, built the report, and did the evidence judging itself in about 15 minutes.
He had a few Windows machines left alongside his Macs, Linux machines and a large NAS. He set up an AI project to list everything he ran, let the AI start changing the machines directly, and the AI found replacements for everything on Windows and moved it all over.
Ported a Linux game binary from Cortex-A7 to Allwinner A133P in about 2 hours using AI, with no source.
Spark read 20 to 30 of the founder's sent emails, extracted his two writing registers, greetings and sign-off patterns, and saved a reusable ghostwriter skill. Days later in a fresh task it drafted a PR email in his media-outreach voice without being told which tone to use.
ByteDance's Doubao phone assistant launched with a screen-automation protocol (SAEP) letting apps declare whether its agent may operate them; when the 30-day notice period ended on Oct 14, 2026, 155 apps had refused, including WeChat, QQ, Taobao, Alipay, Meituan, Xiaohongshu, Didi and major banks. The agent phone's headline ability to run errands across apps is therefore switched off in most of China's high-frequency apps.
He set the AI to export his liked songs from a streaming platform, track 365 artists, and send weekly concert notifications — commenters describe the Saturday roundup as a standing routine.
The AI with Photocraft combined people from two photos of the same location into one photo in 3 tries over 5 minutes, after the AI attempts had changed faces.
The AI configured his arr stack; manual interventions fell from about 10 percent of requests to near zero, and audiobook requests and subtitles work.
He built a VRChat world with GPT-6 Astra and published it. The world is a high rise lounge based on Roppongi, with night views, space to chat, and a game to play.
Making a Pokemon fan trailer in 2 hours with Opus 5.5, a finished artifact.
Vanshika Singhania (2.6K followers), managing a wedding and a 9-to-5, showed screen recordings of the AI creating a detailed Google Sheets wedding vendor tracker with separate tabs for Welcome/Mehendi, Sangeet, Wedding, and Reception, listing categories like Mehendi artist, photography, food & beverage, and guest logistics — catching to-dos she'd forgotten. Another recording shows a weekly content-plan spreadsheet with scheduled video ideas, and the AI attempting a Saturday brunch reservation at Shu
The AI plus GLM did things in his dotfiles folder despite being granted only read access, prompting sandboxing.
Together with AI he ran fairly complex imports from more than 1,000 JSON files into WordPress. Publications and events now arrive as their own custom post type, neatly structured in WordPress, including sideloading the media files.
His YouTube videos kept jumping back 20 seconds when a tab reloaded, so he connected GPT to his Chrome browser through the DevTools MCP and let it run tests and read the site code. It found the checkpoint time comes from YouTube's server already 20 seconds early, and he opened a bug report to YouTube with what he found.
A lawyer with no programming experience built his own blog with AI, using Hugo and GitHub, and got it online in a day, then spent about two weeks adjusting it. He also set up a flow that shares each new post to his Mastodon and Bluesky accounts on its own.
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.