A maker had the AI redesign his KiCad board, including selecting parts from JLCPCB's catalog and suggesting cheaper equivalent components that cut his bill of materials significantly. When assembly proved too expensive, the AI redrew the schematic around through-hole parts he could source himself.
An AI agent named Sarah followed up on her $5 writing pitch by email. He paid, sent edits, and then bought a $3 expansion. Her email read like it came from a freelancer.
A screen-recorded walkthrough of an app called 'Song List' inside an AI app AI shows managing a personal music repertoire: 264 total songs, 76 marked 'In show,' 188 'Needs work,' across 76 artists. The presenter navigates 'In show' and 'All' views and uses the search bar.
Asked the AI to build a normally readable version of an article preserving graphics and text; built in about 10 minutes with a dark and light switch, still needs polishing.
This only partly worked. Worth knowing before you try it.
His wife was left a massive box of coins by her late grandfather many years ago, and they had never catalogued them. He built an app that scans a coin and records its year, type, series and what it is made of, then checks websites for an approximate value and adds it to the collection. It can filter views, track where each coin is physically stored, and show graphs of the total value and the breakdown by coin type. There is also a virtual coin book he can flip through, using the photos he captured while scanning each coin in. For now the app is just for him.
A statistician had argued that class balancing helps mini batch gradient descent. He had the AI Astra design and run a simulation to test the claim, and the results partly corrected him: oversampling sped convergence but hurt probability accuracy, while importance weighted oversampling kept most of the speed and the accuracy. The code was cross checked with AI, and he posted the corrected conclusion as an answer.
The AI updated his system packages, including Node, Python, uv, OpenSSL and certifi to named new versions, checked a checksum, and checked recent vulnerabilities.
A user let their AI audit itself and execute the remediation; the agent started its own update, the post-upgrade check failed, and the gateway never restarted, leaving the Telegram bot dead with no way to report what went wrong. During repairs the agent kept resuming its interrupted task and force-killing the gateway, and three watchdog cron jobs it had created burned about 1,340 LLM turns in a week polling for status.
This one went wrong. See what happened before you try it.
An agent researched available tools, built an eval set, and tested tools against each other to pick the set used for his singing app, cutting build time from months or a year to about 10 days.