Used nit-picking history-teacher agents to evaluate a few hundred generated stories for accuracy, and pointed the AI at HN comments to aggregate issues and fix them with his OK; reported issues mostly fixed.
He could not find an uptime and performance monitor that met his needs, so he built a lightweight one called Blip with help from the AI. He put the code on GitLab and wrote a blog post about it.
He bought a Humble Bundle of Android games 10 years ago, but they are 32-bit and will not run on his 64-bit phone. He asked the AI to decompile and recompile them, and the games work on his phone again.
Austin Evans, a verified tech creator (510K followers), showed his AI 'Lotus' tracking when tickets go live for events such as the United States Grand Prix in Austin, alongside a Collectibles Tracker database and morning tech-news briefs. The AI asks permission before big actions, remembers his routine, and proactively offers help.
He asked the AI to move a long the AI conversation into a coding project. The AI picked a Safari extension he had written himself, recovered 1,651 messages and 97 artifacts, and checked the files.
A creator built a free browser-based video editor with AI for his own clip work after tiring of heavy desktop editors. It finds loops, trims, and exports using in-browser FFmpeg.
Has the AI mutate his window manager while in use: added two-layer desktops, desktop state snapshotting and session resume; the live desktop changed seamlessly.
A genealogist handed his AI an anonymized GEDCOM with about 3,000 individuals. The AI wrote a Python program drawing 300 frames into a 10-second MP4 animated pedigree (ancestors fade in by birth order, glow while alive, dim at death, living shown as 'Living'), then produced an 11x17 300-DPI print from a single frame.
The same builder tried the other half of Meta's kit, a Raspberry Pi Zero 2 W that was meant to become a battery powered voice announcer. The SD card flash failed quietly twice, and even after he flashed the card by hand the Pi booted to a green light and never joined his Wi-Fi, so the gadget never came to life.
This one went wrong. See what happened before you try it.
Joao Silva installed the Jev decision-model skill into his AI fleet without writing integration code: he pasted a plain-English instruction into a session and the agent copied the folder, found the right .env file, added the key and ran the checksum verification. The skill then passed his mandatory three-reviewer gate (security, QA, technical writing) before being committed.
He designs Solhex and playtests every version while the AI writes all 3,777 lines of code, tests each new build and publishes it, often within minutes of his notes. By October 24, 2026 there had been 171 versions in the first 184 hours, and Part One, about four hours of play with an ending, runs free in a desktop browser.
He downloads all his lecture slides and feeds them into NotebookLM for an audio overview. A pipeline then transcribes the audio, has the AI write a timed spec, uses AI Design for the animations and the AI for the filler content, then renders and stitches everything into an hour long recap of the whole semester.
Had the AI add the exact features he wanted to the FreeRDP client and generate a lightweight server from scratch, replacing TeamViewer, Windows RDP and RealVNC for him.
He built Tidehook with GPT-6 Astra and generated art. Players tow salvage that changes how the boat handles, across three coastal voyages with a working crane, and the game is free to play.
A father had the AI build a standard writing practice template for his daughter, who reads well ahead of her level but still writes basic, sound-it-out spelling. The AI also makes the picture for each sheet. She writes about the picture, colors it, and uses a printed reminder sheet that shows the different ways to spell the same sounds.
Sends Nix packaging jobs to an agent and gets a working package, or 20 dependency packages, in about 20 minutes; the AI also fixed his monitor wake-after-sleep config in minutes.
Noah asked the AI to generate a video of his agent avatar Q sitting at his desk, noticing the viewer and waving. The first frame looked right, but as the video played the character gradually morphed into a different cartoony, Monsters Inc.-style figure, and no retry kept the style or character consistent. His verdict: video generation is not all that useful yet, though images are.
This one went wrong. See what happened before you try it.
He needed an app on his QNAP NAS that only existed as open source Linux code, with no NAS package. He gave the AI the model, the AI searched the web and gave him a downloadable package in about ten and a half minutes, and he installed it and it works.
He gave the AI a profile on his Synology NAS, linked over Tailscale by SSH, with one hard rule: never delete anything without his clear say so. The AI went through his large MP3 collection folder by folder. It renamed files, removed duplicates, found album art, fixed tags, and moved songs that were filed wrong, like songs in an 80s folder that were not from the 80s. It also relinked a playlist after he had moved all of its tracks into other folders.
The musician creator recorded a 90 second experimental techno track and generated every shot of its music video locally on MiniMax H3 from a single year old Midjourney image. An Augmentor agent over MCP controlled ComfyUI for the generation, because the creator did not fully know ComfyUI, and the creator edited the result in DaVinci Resolve, building the story around an unexpected second character the AI produced.
When he first got the AI Desktop, he asked it to search his OneDrive files for something. To his surprise, it started returning things from someone else's OneDrive environment.
This one went wrong. See what happened before you try it.
He found a 24 chapter site on CAD history that would once have taken days of fiddling to convert by hand. Working with GPT for a couple of hours, he turned it into a clean offline EPUB for his phone, with linked footnotes and captioned illustrations.