"Don't Make Me Think" in the age of LLMs

We're about to enter a "UX war" between conversational interfaces and old school software design components. For every good product builder, the AI text chat box will be another tool in their toolbox when answering the question they have always asked:

"What helps my user get the job done faster, with the least amount of confusion?"

A common user frustration today goes something like: “This UI isn’t getting quite what I want to accomplish. I’m not sure what these buttons do. I wish I could just explain in plain English.” Conversational interfaces can shine here.

Writing what you want to do mirrors how you think. It's unbounded. The only constraint is your creativity. For tasks that are just easier to explain, you can use your copilot or assistant accessible to you via chat. That's a giant leap forward for user experience. LLMs are the watershed moment for this.

However, when you write or chat, you do have to think! And sometimes you don't want to; you want the software to do the thinking for you.

Deployed incorrectly, new AI interfaces will give rise to a new complaint: “I don’t have the vocabulary to express exactly what I want to do because I’m not even sure myself. Certainly feels awkward when I try. I want it to tell me what to do. I’ll just opt-in to its suggestion.”

The classic web UX book "Don't Make Me Think" from 2000 (!) captures this. It argues that users want to simply opt-in to the very first available solution put in front of them, not slowly deliberate between options at every screen in a funnel.

That sentiment isn't outdated. Humans haven't changed since 2000. For a ton of use cases, constraining the UI with these old school, familiar design elements will help the user get the job done faster than "explaining" things to an LLM.

Strong product makers will resolve this by looking for real UX problems to test their new tool against. The trick is to find areas where the old UI elements fail, evidenced by frequent user frustration. My personal bets so far as an investor have been on categories like: database administration, taxes, and knowledge bases. Certainly many more out there…