AI needs a new UI

Jun 15

As the joke goes, the best place to hide a body is on the 2nd page of google results. No one clicks to the next page.

The Google search paradigm

Google built the paradigm for all of search around an extremely simple user experience: user asks a simple question (search query) and Google serves up relevant results in the first few entries. A design principle of Google early on was that they would not ask anything more from the user. This was revolutionary–I remember earlier search engines such as Lycos, Alta Vista, Yahoo employing various filtering tools as part of the experience. Google’s genius was doing the opposite: they focused all of the improvements on delivering better results while explicitly never asking more of the user.

Text entry field, submit button. User types a question. Google provides an answer. This interaction became the most ubiquitous digital experience on the internet. Everyone has used it thousands of times. The pattern is deeply engrained in our minds. Any deviation feels wrong.

The AI interface mismatch

And now AI comes along, gives us the exact same UI, and expects us to behave differently. In that familiar text field, a user must write a prompt that has several parts to provide context, persona, desired output format, and more. Then they’re expected to have a back–and–forth conversation with the interface to refine the output and provide additional context.

You see the problem?

The interface promises the simplicity of search but demands the complexity of programming. We’ve trained users for 25 years to expect immediate, comprehensive results from brief queries, then asked them to master an entirely different set of skills using an identical user interface.

Early adopters and tech enthusiasts will endure this learning curve–the pain of mastering something new is worthwhile when they can envision the potential. But that’s not most people.

I’ve been using AI for over a year. Initially it was frustrating. I believed the hype but couldn’t realize the practical (promised?) value. Only through persistent experimentation did I learn effective prompting techniques and identify valuable–to me–use cases. The technology evolved from awkward and forced to genuinely additive. But the key point: it required significant effort on my part to get there.

From one interface, infinite outcomes

Perhaps the biggest user challenge is that people can’t grasp the breadth of what AI can accomplish. Consider these wildly different use cases:

Blog post: Generate content based on a simple theme or a couple lines of an original idea.
Mockup: Generate a clickable prototype of a mobile app based on simple instructions.
Training program: Develop a schedule with routine steps to follow so that I can develop a new habit.
Presentation: Create an outline for a slide deck presentation.
Debate: Help someone think through a complex idea and identify gaps and opportunities.
Visual ID: Generate a whole brand identity for a new product.

Each started from the same UI: a blank text field.

The possibilities feel endless, which paradoxically makes AI feel overwhelming and impractical. Without effective signposts, users can’t keep track of all the many things AI might accomplish for them. The cognitive load of remembering and formulating requests for dozens of potential use cases creates a massive adoption barrier.

Building a new interface paradigm

What should replace the simple search box? While the complete solution remains unclear, several key principles emerge:

Contextual awareness. AI doesn’t need to go in every direction when I first engage it. It should understand who I am, where I am, and what I’m likely trying to accomplish, then adapt to support the most probable use cases. Memory of past interactions was a step-change improvement when LLMs began incorporating conversation history. This needs to go deeper.

Adaptive UI. Because possible outputs vary greatly, the inputs will need to be equally different. Sometimes text is optimal. Other times voice, images, or autonomous execution make sense. The UI should morph to match the task rather than forcing everything through a single input method.

Discrete task-focused components. Rather than cramming infinite possibilities into one box, we need specialized agents that excel at specific functions. These components should understand each other’s function and connect intelligently. When users have a particular need, the appropriate agents should collaborate to deliver an optimal result.

Real-time feedback and guidance. Current AI interfaces leave users guessing whether their approach will work. The system should provide insight into what’s working in a prompt and suggest improvements, rather than leaving the user to experiment blindly.

Refinement tools. One of the biggest frustrations with AI is a “close-but-not-right” output. It is really hard to make small but important adjustments to an AI output. This is particularly true with image creation or generated prototypes. More precise tooling around editing will be immensely helpful and avoid needlessly starting over.

The path forward

The companies that solve this UI challenge will be doing more than just improving the user experience–they’ll be unlocking AI’s mainstream potential. It isn’t about incremental improvement to chat interfaces. It’s about fundamentally reimagining how humans and AI collaborate.

Google’s breakthrough was radical simplification–removing friction that competitors had normalized. AI needs its own paradigm shift. We need interfaces that wrangle the inherent complexity of AI into something that feels effortless and intuitive.

AI will transform how we work, create, and solve problems. However, this transformation will remain beyond most people until we redefine the interface in a way that makes sense to humans.

Featured-AIAI

Ken Sigel