Build it with AI or just buy it? An honest way to decide

"Why would I buy that? I'll just ask AI to write it." It's a fair reaction, and often it's the right call. But the comparison people make is the wrong one: they compare the price of a script to zero. The real comparison is the price of a script to the cost of your time — plus the weeks afterwards.

What AI is genuinely great at

Give a model a well-specified, self-contained task and it will produce good code faster than you can type. Renaming files by pattern, reshaping a CSV, a one-off API call, glue between two systems you already understand — write those. Don't buy them.

Where the estimate quietly falls apart

The trouble is that AI produces something that looks finished long before it is. The first version runs on the happy path. Then reality arrives:

• The API paginates, and the code only fetched page one.
• Rate limits exist, and nothing retries.
• Auth token expires after an hour with no refresh.
• It works on your machine because of a dependency you forgot you had.
• Nobody handles the empty result, and it writes empty rows for a week.

None of these are hard. They're just unknown until they bite — and you can't prompt your way around a failure mode you haven't met yet. That's where "twenty minutes" turns into two days.

The test

Ask two questions:

1. 1Is this specific to my business, or has it been built thousands of times before?
2. 2Do I want to own it in six months — the fixes, the breakages, the edge cases?

Two yeses: build it. It's your logic and you should own it. Two nos: buy it. You're re-solving a solved problem and then inheriting its maintenance for nothing. One of each: usually buy, then adapt.

Buying isn't the opposite of building

This is what gets missed. You're not choosing between your code and someone else's black box. You get the full source. The realistic path is: buy something that already handles the boring 80%, then use AI to reshape the last 20% to your case — with working code as context instead of a blank file.

AI is much better at editing something that works than at inventing something that has to work.

So the honest answer isn't "buy, don't build." It's: don't spend three days rediscovering failure modes someone already fixed. Spend them on the part that's actually yours.