We use AI in our work. We also tell clients exactly where, and exactly where we do not. That should be unremarkable. The fact that it counts as a differentiator says something uncomfortable about how the rest of the industry is behaving.
The rule we work to
It is simple. Nothing runs unchecked, and nothing gets built that the client cannot understand well enough to own. AI is genuinely good at some things: the copying between systems, the first draft, the boilerplate code, the sorting and tagging that eats a week. It is not a reason to stop looking, and it is not a substitute for judgement.
So when we automate something, we tell you what the machine is doing, what it is not doing, and where a human still needs to be in the loop. If a piece of writing or code was produced with AI, we say so. You should never be surprised later by how something was made.
Why the honesty matters
Two reasons. The first is trust. If you cannot see where the AI is, you cannot judge whether to trust the output, and you certainly cannot maintain it after we have gone. Hidden automation is a liability dressed as a shortcut.
The second is quality. AI is confident even when it is wrong. It invents sources, misreads context, and smooths over the exact details that matter. Left unchecked, it produces work that looks finished and falls apart on contact with reality. Checked properly, by someone who knows the subject, it saves real time. The difference is the human, not the model.
The honest version
The honest version of AI is less exciting than the sales version. No promise to multiply your output overnight, no pretending a tool can replace expertise it does not have. It is a good assistant and a poor decision-maker, and we treat it as exactly that. That is the version that still works in a year, once the hype has moved on.