A working guide, September 2026

Using AI to build a side business

What actually works right now, written by someone doing it rather than someone selling a course about it. No prompt packs, no ten-thousand-dollar-a-month promises — the tools I use, what they cost, what they are genuinely good at, and the parts where they waste your afternoon.

PDF · about 7 pages · 2,983 words · read it on anything

What's in it

  1. Before anything else

    I am going to tell you what I actually do, and what it actually costs, and where it actually falls over.

  2. Part one: the mistake nearly everyone makes first

    The instinct is to ask what AI can do, and then go looking for a business in the answer.

  3. Part two: what I actually pay for

    Prices as of September 2026.

  4. Part three: getting from idea to something a stranger can pay for

    The order matters more than any individual step, and the order is not the one that feels natural.

  5. Part four: how to actually work with a coding agent

    This is the part where most people leave value on the table, so it gets the most space.

  6. Part five: the five failures, from someone they happened to

    The most common one and the most expensive.

  7. Part six: a realistic first month

    Not a promise.

  8. Part seven: what I would do if I started over tomorrow

    1.

  9. A closing note, and the honest part

    The advantage available right now is real, and it is temporary in a specific way: it is not that these tools exist — everyone has them.

Read a section first

This is one section of the guide, exactly as it appears in the PDF. Prices and all — nothing is held back for the download.

What I actually pay for

A coding agent, at the top tier. This is the whole game and the only line item I would tell you not to economise on. Around $100–200 a month for the tier that does real work; the cheap tiers exist and will frustrate you into quitting.

I use Claude Code. Cursor and Codex are the main alternatives and people do good work in all three — the choice matters much less than the tier you buy. Pick one, buy the level that does not make you wait, and stop reading comparisons.

Why this one is non-negotiable: at the good tier, you describe a feature and get a working, tested implementation. At the cheap tier, you get something that nearly works, and you spend the evening finding out how. The cheap tier is more expensive.

What it isn't

  • It is not a course, and there is no upsell at the end of it.
  • It does not tell you how much money you will make, because I do not know anything about your situation and neither does anybody else who says that to a stranger.
  • It has no prompt pack in it. The good models stopped needing incantations and selling you a list of them would be taking your money.
  • It goes stale. It is dated on the cover for that reason. If you are reading this a year from now, the tools section is wrong and you should ignore it.