Build a business with AI — and prove you can.

What you can do afterwards

1. Narrow a vague idea into a wedge: one specific person, one recurring problem, one budget that already exists.
2. Say in plain words what a model can and cannot do — enough to scope a job and answer a client without bluffing.
3. Turn a service into a product: one outcome, one price, one written process — and explain why “it depends” is unsellable.
4. Run a discovery conversation that qualifies instead of pitching, say the price out loud and hold the pause.
5. Set the money side up so it survives year one: separate account, invoices that get paid, a reserve for the tax nobody has asked for yet.

Who this is for

• You are starting from zero, usually next to a job, and want a first paid client rather than a business plan.
• You use AI tools already but cannot yet say what someone should pay you for.
• You want the boring parts too — pricing, contracts, tax reserves — not just the exciting ones.

The story it is taught in: The whole curriculum follows Jordan Reyes, an operations manager at a logistics firm who builds an AI-assisted service business in the evenings. No overnight success, no quitting the job in week two — realistic timelines and the setbacks that come with them.

What you can do afterwards

1. Describe each of the seven models in one sentence a stranger understands — and name what each one rules out.
2. Compute cost per delivery and find the two steps that actually carry the margin.
3. Recognise the wrapper trap: a tool whose entire value is one prompt has the half-life of one model release.
4. Run the utilisation arithmetic that decides an agency: billable hours, the hours you will never bill, and when hiring beats working longer.
5. Sell to a local business in a language that has nothing to do with AI: fewer missed calls, faster quotes, evenings back.

Who this is for

• You have decided to build something and now need to choose the shape it takes.
• You keep hearing that one model is the answer and want the arithmetic instead of the enthusiasm.
• You already earn something and are wondering whether a second model belongs next to it.

The story it is taught in: Sam Okafor works through the models one at a time, taking each far enough to price it honestly — including the two that turn out not to fit.

What you can do afterwards

1. Explain why models invent — as a property of how they work, not a bug awaiting a patch.
2. Write an instruction that holds and force output structure you can check by machine instead of by hope.
3. Steer an image deliberately — and name the hard limits (text in images, exact product likeness, characters across a series).
4. Compare generated video against a shoot on control, consistency and total time including revisions.
5. Cost a whole job rather than a single call, and find the step that dominates the bill.

Who this is for

• You deliver AI work for clients and need to promise only what the models actually do.
• You are a designer, producer or freelancer and want to know what is steerable and what is legally yours.
• You quote jobs and keep guessing at the cost per delivery.

The story it is taught in: Priya Raman takes one job per model family and learns each by delivering it — a small studio, real deadlines, no agency fantasy.

What you can do afterwards

1. Separate the three things the market calls “agent”: a chat assistant, an automation, and an agent that plans, calls tools and loops.
2. Judge a task against four questions — repetitive, judged, reversible, observable — and say no out loud when it fails them.
3. Build a working assistant in an afternoon without a server, and name the ceiling of that stage.
4. Handle what nobody demos: retries, partial failure, duplicate runs, and a ceiling that stops a loop from becoming an invoice.
5. Scope a client engagement so “done” is testable: named inputs, named outputs, an acceptance list, a written non-goal.

Who this is for

• Your boss or your client said “build us an agent” and you need to know what that means before you agree.
• You have automated something once and it broke in a way nobody noticed for two weeks.
• You want to sell agent projects and need a scope that does not end in an open-ended argument.

The story it is taught in: Noor Haddad automates their own team’s work first and only then sells the capability — evenings and weekends, no engineering team.