Where Friction Belongs
I have a love-hate relationship with friction.
When I was first learning how to use GenAI tools, friction was helpful. It forced me to slow down, check my work, and remember that a polished answer is not the same thing as a trustworthy one. At that stage, friction was not getting in my way. It was protecting me from my own inexperience.
That changed once I understood the pitfalls better. Once you know what to verify, where the system tends to mislead, and how easily confidence can outrun accuracy, friction starts to feel different. It slows you down, breaks your rhythm, and makes a useful tool feel more cumbersome than it needs to be.
Now, as these tools become more capable, seamless, and persuasive, I can already see that some of the friction I was happy to lose may need to come back.
A lot of legal and judicial technology is being sold right now on the promise of a frictionless workflow. To be fair, some friction should go. Duplicate entries, disconnected systems, clumsy interfaces, and pointless administrative burdens do not protect fairness or accuracy just because they have been around for a long time. That kind of friction is not a safeguard. It is just drag.
So I am not opposed to frictionless design. In fact, I think that is exactly where we should start.
Design the workflow frictionless by stripping away all of the dead weight. Let the system organize, summarize, draft, route, and surface information with as little unnecessary resistance as possible. Build the smoothest version you can.
But then ask where the friction needs to go back in, and why.
The answer depends on the user, the task, and the capability of the tool. As the tools become more capable, even experienced users may need new friction added back in.
What concerns me right now is not just that a strong AI system can be wrong. It is that it can be wrong in a way that feels effortless. The answer arrives polished, the workflow feels smooth, and the system seems helpful. That makes it easy to miss the omitted fact, the unsupported inference, or the moment when the machine quietly starts shaping judgment instead of merely assisting it.
That is where people get into trouble.
For the justice system, the question is not whether friction is good or bad. The question is where it belongs. Where should the system move quickly, and where should it force a pause? Where should it reduce burden, and where should it require visibility, review, and human ownership?
The justice system does not need friction everywhere. It only needs friction in the places where it still serves a purpose.
Design it frictionless. Then put the friction back where it belongs.
