AI in a day: what actually fits

What can you actually build with AI in a day?

One working thing, doing one job. A landing page that takes real signups, a single automation that removes a task you repeat every week, one agent that answers one type of question, or a decision you have been postponing for a month. What does not fit is a product: accounts, billing, and the long tail of exceptions are week-three work, not day-one work.

That is not a guess. A landing page with a working signup form went from an empty folder to a deploy-ready build in one day, including breaking the form and fixing it twice. An agent that routes customer questions took an afternoon and got 19 of 20 test questions right, with the one messy case flagged for a human instead of guessed wrong. A to-do list of twelve tasks took a morning and came back as four.

Every one-day project, shortest first

Here is the full index of projects that ran against a measured day. The time column is what it actually took, and the last column is where each one ended up, including the ones that ended badly.

What does not fit in a day?

Anything that touches money, other people's data, or a long tail of exceptions. Those projects are in the index too, because the failures are the useful part.

• A freelance CRM stalled in an afternoon. The dedicated tool was not the answer, the $0 spreadsheet nearly was, and it still broke somewhere.
• An AI browser agent handled a day of errands, but every money step stayed in human hands on purpose.
• Five AI writing tools got one brief to write a bio. None of them got it right, and each one broke differently.

There is a pattern in those three. AI closes the distance to a first working version fast, then the last stretch, the part where something has to be correct rather than plausible, still costs you the same hours it always did. Plan the day around the first version and you finish. Plan it around the finished thing and you do not.

Is this the same as an “AI in a Day” workshop?

No. Most results for that phrase are one-day training sessions, where a company or campus books a room and a trainer explains what AI is. This is the other kind of one-day: a library of projects you can copy tonight, with the tools, the prices we paid, and the parts that failed left in. No classroom, no signup, no certificate.

How to spend your next day

Pick the smallest version of the thing you want, and give yourself until the end of the day. Four steps that hold up across every project in the index:

1. Choose a task you already understand, so you can tell a good result from a plausible one.
2. Write the brief before you open a tool. One paragraph, with what “done” means in it.
3. Build the ugly working version first. Styling and edge cases come after it runs.
4. Stop at the end of the day and write down what broke. That list is your second day.

If you want a starting point instead of a blank page, the four pillars below each hold a day's worth of work, and the tools index shows which tool showed up in which project.