It's easy to bolt AI onto anything. Slap a chatbot in the corner, add a 'generate' button, call it innovation. It's much harder to build one that removes work instead of adding a new thing you have to babysit.
Our rule is boring but it holds up: a feature has to earn its keep. If it doesn't save real time or catch real mistakes, it's a sticker on the box, and we'll say so before we build it rather than after you've paid for it.
The failure mode is everywhere. A chatbot that can't actually answer anything and just funnels people to the contact form they were already looking for. A 'smart' summary you have to re-read the original to trust. Features that demo beautifully and then sit unused because, in the actual flow of the day, they're more hassle than doing it by hand.
The honest truth is that a lot of what small businesses need isn't AI at all. It's a fast website, a booking form that works, a list you can edit. Reaching for a large language model to solve that is like buying a forklift to carry the shopping. We'd rather build the plain thing that works than the clever thing that impresses.
But when it does fit, it's genuinely transformative - and the tell is always the same: it takes a job a human currently does over and over, and does it well enough that they stop doing it.
The clearest example we've built is Emily, a voice assistant that understands a whole business and actually acts - you ask in plain language and she does the thing, hands-free, instead of you clicking through five screens. That earns its keep every day, because it removes work that used to have a person's name on it.
So we use AI heavily where it helps - including in how we build sites this fast - and we leave it out where it's decoration. The question is never 'can we add AI here'. It's 'if we do, what stops being someone's problem'. If there's no good answer, we don't add it.