An AI in the room

Every project here starts with two people in the room, and one of them isn’t a person. We put an AI model into the same conversation where we plan a feature, sketch a screen, or debug something at 11pm. Not as a tool we occasionally consult. As a participant in the thinking.

That sentence makes some clients nervous, so let’s be specific about what it means and what it does not.

It drafts, it doesn’t decide

The model writes the first version of almost everything: the component, the migration, the email copy, the test cases nobody wanted to write by hand. It is fast in a way no team of people can match, and it doesn’t get tired of the fortieth variation of a form.

What it does not do is decide what we’re building or why. Scope, priorities, and the calls that actually cost money if we get them wrong stay with a person. When a model suggests an approach, someone here has to be able to say why it’s right, not just that it compiled.

The room gets bigger, not quieter

The obvious fear is that adding a model to the process means fewer humans doing the thinking. We’ve found the opposite. Because drafting is faster, we spend more of our own time on the parts that used to get skipped: reading the actual support tickets before designing the fix, sitting with a client’s spreadsheet instead of guessing at their workflow, testing the ugly edge case instead of shipping the happy path.

The AI didn’t shrink the job. It moved where the hours go. Less time typing out boilerplate, more time on the decisions boilerplate can’t make for you.

What we watch for

Two failure modes show up if you’re not careful. The first is a model producing something that looks finished and isn’t: correct syntax, wrong assumption. We catch this by reviewing everything the way we’d review a junior engineer’s pull request, not the way we’d rubber-stamp our own work.

The second is slower and sneakier: a team that stops building the judgment a model can’t have, because the model always has an answer ready. We fight that by making sure a person always makes the call that matters, even when the model’s draft would have been fine. The muscle atrophies if you never use it.

Why this is the whole company

This isn’t a policy we adopted. It’s the reason Nodo exists. Two nodes, one human and one model, and the interesting work happens in the overlap: where the model’s speed meets someone’s judgment about what actually matters. Neither one alone gets you there.

If you want to see it in practice, our case studies describe exactly who decided what, and where the model’s draft ended and a person’s decision began. That’s not a footnote for us. It’s the entire pitch.