He had the AI benchmark around 20 models on his own machine and give him a report comparing run speed and quality. After 7 hours of testing it said gpt-oss-20b was the best fit for his setup, and that is the one he chose.
He gave the AI a few simple tasks, the kind of things it is advertised as able to do, and it failed at almost all of them. It ended up costing him much more time than it would have saved.
This one went wrong. See what happened before you try it.
The AI autonomously ran Quake demo playbacks, took pixel-diff snapshots, compiled a native build and patched the game for determinism, leaving a solid regression test set.
He replaced Evernote with Apple Notes. The one thing in the way was his old document scanner, which would not scan straight into Notes, and the AI solved that problem fairly quickly.
Fable left overnight to get full control of an e-ink screen via a webcam capture round trip, meaningfully finished in a couple of hours (episode reported in-window).
A user asked the AI to download a data export from five one-time links. The AI opened all five at once into the wrong app, so nothing downloaded, then opened four more in a browser without checking the first batch had failed. About ten windows and tabs later the one-time links may be used up. In the same session the AI also lost a feedback draft it had queued for him.
This one went wrong. See what happened before you try it.
She uploaded the first 10 images by hand, adding each prompt as alt text. Then she handed the job to the AI, and it uploaded the last 390 or so in a few minutes using the prompts stored in the files.
The AI fixed Steam games and HiDPI and monitor issues on his Linux machine and powers a recurring shell script that checks updates and looks online for potential issues.
Creator took on a $30/week family-of-4 budget challenge with AI, asking it to start from what she already had in her pantry. The AI produced 7 dinner ideas, a grocery list, and a price comparison; her Walmart haul came to $29.66, under budget.