Good Products Remove Work, Not Judgment

A common product ambition is to reduce the number of steps between intent and result. Fewer fields, fewer clicks, fewer decisions. With AI, we can take that ambition much further: the product can interpret the user’s intent and complete much of the journey on their behalf.

I don’t think that automatically creates a better product.

Some friction is simply waste. Asking customers to enter information we already have is waste. Making them copy data between systems is waste. Requiring an internal approval because our permissions model is too crude is also waste. Product teams should remove these things without hesitation.

But not every action is an obstacle. Some actions help users understand what they want, make a deliberate choice or accept a trade-off. Automating those actions can make the experience look smoother while making the product less useful.

Consider a campaign tool that can generate an audience, write the message, select the channel and schedule delivery from one short prompt. That is impressive when judged by time to launch. But speed is not the only relevant outcome. Does the user understand why that audience was selected? Have they considered which customers should not receive the message? Do they know what success will look like? If the product skips these decisions, it may shorten setup while increasing the chance of a poor campaign.

The same issue appears in onboarding. Teams often try to reduce onboarding to a single question and let the system configure everything else. That can work when the user’s needs are predictable and mistakes are easy to reverse. It works less well when configuration reflects company policy, commercial priorities or risk tolerance. In those cases, asking the user to make a few explicit choices is not bad design. It is how the product establishes shared understanding.

I use a simple distinction: administrative friction versus judgment friction.

Administrative friction is work the system should do. Finding records, formatting content, checking for missing information and applying known rules are good candidates for automation. Judgment friction is the moment where someone needs to choose between meaningful alternatives. The system can support that decision, but it should be careful about making the decision disappear.

This matters for growth as well. A shorter funnel can improve completion while weakening commitment. If users arrive at an outcome without understanding how it was produced, they may be less likely to trust it, repeat it or explain it to a colleague. That creates an attractive activation chart and a weak habit.

I would rather measure whether users reach a valuable result and can confidently repeat it. That may require an extra review screen, a comparison between options or a clear confirmation of what the system is about to do. Those steps are justified when they improve the quality of the decision, not merely because stakeholders prefer another checkpoint.

There is also a discovery risk. User choices generate information. When people adjust a recommendation, reject a proposed audience or change a generated plan, they tell us where our assumptions are wrong. If we automate the entire flow and only measure completion, we lose that signal. The product becomes easier to use and harder to learn from.

Before removing a step, I want the team to ask what function it performs. Is it transferring data, satisfying an outdated process or compensating for a technical limitation? Remove it. Is it helping the user understand consequences, express a preference or take responsibility for a decision? Improve it, but think twice before eliminating it.

The goal is not a product with the fewest possible interactions. The goal is a product that spends the user’s attention well. AI should reduce the work around a decision. It should not quietly remove the decision itself.