AI is infrastructure, not the product
The most useful thing artificial intelligence can do is disappear. Why we build AI into every product and market it in none of them.
There's a reliable way to tell whether a company has thought carefully about artificial intelligence: check whether AI is the noun in their pitch.
"AI-powered." "AI-driven." "AI for X." When intelligence is the headline, it's usually because the underlying system doesn't have one of its own. The technology is being showcased because there is nothing else to show.
We take the opposite position. AI runs through everything we build, and it is the product in nothing we build.
Electricity, not fireworks
Nobody markets a building by celebrating its wiring. Electricity is everywhere in the building and visible nowhere — and that invisibility is precisely what makes it useful. You flip a switch; light happens.
That's the standard intelligence should be held to inside software. It should:
- understand context without demanding configuration
- remove work before adding features
- reduce the number of decisions a person has to make
- surface a recommendation when one is warranted
- stay out of the way the rest of the time
When AI meets that standard, users don't say "what impressive AI." They say something far more valuable: "this was easier than I expected."
Automation removes work. Intelligence removes decisions.
The distinction matters more than it sounds. Automation takes a task you already understand and does it faster. Intelligence takes a decision you were carrying and carries it for you — or reduces it to a confirmation.
Most people running a business aren't drowning in tasks. They're drowning in decisions: what to post, how to price, which form to file, what to do next. A system that automates tasks while leaving every decision in place has helped less than it appears.
This is why we prioritize cognitive load over horsepower. The question is never "what can the model do?" It's "what can the person now stop thinking about?"
Agency is not negotiable
Reducing decisions is not the same as removing judgment. Intelligence in our systems recommends; it doesn't quietly decide things a person would want to know about. It shows its reasoning where reasoning matters. It never pretends certainty it doesn't have.
The goal is a person who is more capable and more independent — not one who is dependent on a system they don't understand. If using the product makes you feel less in command of your own business, the system has failed, whatever its benchmarks say.
Why this matters now
Models will keep improving on their own. What won't improve on its own is the layer between the model and the person — the system that decides which problems are worth solving, which decisions can be absorbed, and which must stay human.
That layer is design work. It's the work we consider ours.