Build versus buy with AI

Now that building software is easier, I keep coming back to the same question: when should I build something myself versus pay for an existing product?

One question I’d start with: how much feedback will it take to make it good?

If I can describe exactly what I want and tell immediately whether it works, a few prompts might get me there. A script that turns one file format into another is a good example.

If the hard part is figuring out the right UX, I still need to use it, get annoyed, change it, and repeat. An existing product may already have gone through years of that with its users. Recreating its visible features doesn’t tell me how much of that learning I’ll need to do myself.

This is where feedback loops affect the decision. AI can help me try an interface quickly. Some of the feedback I need will only arrive after I’ve lived with it for a while.

A few other questions I’d ask:

• How much of the product do I actually need? Replacing one feature is a different project from replacing the whole thing.
• What happens when it breaks? A personal tool I can fix tomorrow is different from something my team relies on every day.
• Will I still want to maintain it in six months, when the fun part is over?
• Does an existing product force enough workarounds that building my own would actually save attention?

Sometimes I’d just build the missing piece on top of a product I already use. That lets me keep the parts that work while changing the part that keeps getting in my way.

I’m still figuring out where to draw the line. How clearly I can describe the tool is a useful starting point. How much I’ll need to learn by using it may be harder to estimate.