AI should automate the work users hate, not the decisions they care about.

There are two ways to put AI into a product. One removes the work people dread. The other removes the decisions people care about. They look similar in a demo and could not be more different in practice.

The first kind is quietly transformative. It drafts the thing nobody wanted to write. It fills in the tedious fields. It reconciles the numbers. Then it hands control back to a person who reviews, adjusts, and decides. The person feels like the product did something for them.

The second kind is where most AI theatre lives. It makes the choice, hides the reasoning, and asks the user to trust it. When it is right, it is unremarkable. When it is wrong, it is expensive, and the user had no way to catch it. The person feels like the product did something to them.

A simple test

Before adding AI to a workflow, ask which side of that line it falls on. Is it removing labour or removing judgment? Labour is a gift. Judgment, taken without consent, is a liability.

This is not a case against ambitious AI. It is a case for putting the ambition in the right place. Draft aggressively. Decide conservatively. Keep the human as the author, even when the machine does most of the typing.

Why it matters commercially

Products that automate the work users hate get adopted, because they respect the user. Products that automate the decisions users care about get abandoned the first time they are confidently wrong. Trust is the scarce resource, and it is spent the moment a person loses control of something they are accountable for.

The best AI I have built and advised on shares one trait: it is fast at the work, and patient with the decision.

Written by Ryan Short. Start a conversation or read more on LinkedIn.