The benefits and the downsides don't cancel out: they compound

In the last post: is this actually the right problem to build software for? Say the answer is genuinely yes: there's a real gap, no existing tool fills it, it's worth the investment. The next question is just as important: built how, and by whom, and with what happens after launch already accounted for.

AI-assisted development has a clear upside column: speed, lower upfront cost, the ability to try three approaches in the time it used to take to plan one. For a small team without a large budget, that can be the difference between shipping something and shipping nothing.

The downside column is less visible, which is exactly the problem:

• Requirements that were never fully written down, guessed at, and now baked into code that looks finished.
• No single person who can explain why the system is structured the way it is.
• Security and data-handling decisions made implicitly, by whatever the tool defaulted to, rather than explicitly, by someone accountable for the choice.
• Technical debt that doesn't announce itself: the system works fine right up until it needs to do something slightly different, and every shortcut becomes a wall.

These two columns aren't separate. They compound, over the life of the product. Speed today is very often the direct cause of fragility later, not because AI-generated code is inherently worse, but because the deliberate steps that catch problems early are the exact steps that get skipped when speed is the whole point. A "good enough for now" MVP has a way of quietly becoming the permanent system of record.

None of this means don't use AI tools to build: it's a large part of how we build too, including AI-powered functionality itself when a project calls for it. It means the tools are excellent at execution and indifferent to judgment. A short, deliberate diagnostic step before development starts is usually all it takes to shift most of your outcomes from the downside column into the upside one.

That diagnostic step is the part of our work we probably talk about least, and it's the part that matters most. More on that next.