There is a version of the AI conversation that is entirely about subtraction. How many people, how many hours, how much cost. It is not wrong, exactly. It is just the least interesting question available.
What is AI actually for inside a company?
Reach. The person in accounting who understands the filing system better than anyone but has never written a query. The service agent who knows why a process breaks but has no way to fix it. The designer who can see the system but not build it. Every organisation is full of people who are one capability short of solving their own problem.
That is the gap worth closing. Not headcount, capability. A system that lets someone act on what they already know is worth more than one that acts instead of them.
And the part machines do not do
I design with my hands. Still, after sixteen years and after every tool that was supposed to make that unnecessary.
AI will solve the task. It will produce the layout, the copy, the schema, the component. What it does not produce is the reason one version is right and the other is merely correct. That judgement is the whole job. The signature of a designer is not the output. It is the decision about which output is worth having.
So the systems I build assume a person at the end of them. Confidence figures instead of confident answers. Sources you can check. A place where someone says yes before anything is final.
Not because the machine is untrustworthy. Because the work belongs to the person, and a system that forgets that produces things nobody wants to sign.