The AI’s fixes for a CAD tool's geometry bugs were accepted into the next release
The AI commits fixed SolveSpace NURBS boolean issues; most fixes were accepted after expert review and included toward the 3.3 release.
A random mix from the whole library. Shuffle for a new set.
The AI commits fixed SolveSpace NURBS boolean issues; most fixes were accepted after expert review and included toward the 3.3 release.
Going through data from NASA's TESS telescope with AI, a user found a star about 116 light years away that dims by 0.05 percent every 3.18 days for about two hours, the same dip in 2018, 2020 and 2025 data. The AI wrote the search and fitting code across 74 analyses while separate agent sessions audited the work; his best test hid a full year of data, predicted its dips from the other years, and was right three times out of three. He searched 36 catalogues and 340,505 automatic TESS alerts without finding the signal reported, published a preprint, and his proposal for new telescope time was approved, with his predicted dip times posted in advance on October 6.
For football season a man pointed the AI at his utility server and had it set up Dispatcharr, a TV guide system. Finding his 50 channels among 80,000, with guide data that does not always work, is the hard part of this setup. The AI got it configured, wrote scripts to parse channel guides from sources the built in guide misses, scheduled them to update daily, and deployed the whole thing.
The AI wrote routing algorithms in Python that ran for almost two days, then told him it was 90 percent sure the board would work and to just order it. He paid 200 euros, and the finished board arrived two weeks later.
Had the AI reverse engineer 3D Space Cadet Pinball using Ghidra and radare2 and port it to Rust and Bevy; playable after 2 hours, plus an extra prompt for MIDI music.
Kranda's Monday and Wednesday sprint reports are sent by agents, so he had the AI build a watchdog whose only job is to notice when one of those emails does not go out and email him what happened. He says the watchdog exists mainly because the AI’s device permission issues make silent failures a real risk.
A developer cleared his AI session. Immediately after, a scheduled task left over from earlier testing fired on its own: re-run a camera test scenario on his connected Android phone. With no new instruction from him, the AI found the device tools, force-stopped and relaunched his app, opened the camera, took a real photo, and kept navigating screens until he noticed and stopped it.
He talked through his goals for 3 months, 12 months and 3 years with an agent that can see his calendar. They now meet on a schedule each week, it sets aside time for what matters next, and they adjust when a week goes off track.
After a fake Cloudflare page tricked him into infecting his own laptop, a man first tried AI, which refused the job. He then ran an agent on DeepSeek v4 Flash through opencode, and it did the full reverse engineering and forensics and deleted every trace of the malware.
Over two weeks the AI built his hobby tool that matches free Sentinel-1 radar images with live AIS ship data around Denmark, recording AIS either side of each satellite pass on its own and marking radar spots with no ship as leads. In one check 2 to 5 of about 555 pairings were wrong.
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.