Learning AI by actually using it
Twenty Dollar AI is an independent project run by an everyday AI enthusiast who is learning by doing, not by an AI lab or a company.
The question
Most AI coverage is either breathless launch hype or deep technical benchmarking. This project asks something simpler: what can an ordinary person actually get done with an affordable AI subscription? The name comes from the price range this notebook focuses on: AI subscriptions around $20 a month.
How it works
Everything here is learning in public. I pick real tasks, run them with the tools, and write down what happened, including the parts that didn't work.
- Blog: first-person notes on learning, building, and deciding whether a subscription earns its keep.
- Benchmarks: one real task per run, with measured results and the output.
- Agent experiments: real assignments given to AI agents, with time, usage, outcome, and limitations.
- Projects: things built along the way, linked to public code.
The rules this site follows
- No invented results. Scores, timings, usage, and outcomes come from actual runs, or they don't appear.
- Failures stay in. A test that went badly is still a result.
- Measured and opinion are labeled. On benchmark and experiment pages, recorded numbers carry a Measured label and personal judgments carry an Opinion label.
- Prices need a source. A subscription price is only shown with where it came from and when it was checked.
- Corrections are visible. Updated pages show an updated date.
Follow along
Updates go out on X at @TwentyDollarAI (opens in a new tab), code lives on GitHub (opens in a new tab), and there's an RSS feed for new posts.