Run a model in production. Measure it. Keep it inside a budget.
Assumes Intro to AI for Business. Start here anyway if you like.
Every class teaches one idea, asks you three questions on it and then puts it to work on a real company.
Real inputs. Known answers. A score. Run it every time anything changes.
Cheapest model that passes. Per task. Rerun the eval when anything changes.
Speed. Cost. Quality. Pick two per feature. Fix length before you fix the model.
The prompt is the product. Put it in the repo. Run the eval on every change.
Number every version. Log it on every call. Canary before everyone. Rollback in a minute.
Five numbers a week. Errors predict complaints. Refresh the golden set when the world changes.
Stop the radius. Find the version. Roll back. Write it up. Add the cases.
Send three fields. Strip what identifies. Read the retention terms. Log the call.
One file talks to the model. Test a second vendor every quarter. Read the deprecation email.
The model is a tenth of the bill. Budget the people. Set the daily alert before launch.
Five numbers. Three decisions. Something gets cut or the review isn't working.
Build two by hand. Extract the platform. Intake stops the half that shouldn't be built.
Figures. Model prices from the provider's pricing page read on Sep 7 2026. Everything else is worked arithmetic on stated assumptions.
| Source | Figure | Value | Date read |
|---|---|---|---|
| Anthropic pricing page | Claude Haiku 4.5 | $1 in. $5 out. Per million tokens | Sep 7 2026 |
| Anthropic pricing page | Claude Sonnet 5 | $2 in. $10 out | Sep 7 2026 |
| Anthropic pricing page | Claude Opus 5 | $5 in. $25 out | Sep 7 2026 |
| Anthropic pricing page | Batch discount | 50% | Sep 7 2026 |
Ten minutes. No account. No card.
Take my first class