Study live products the way we used to study interface patterns. The box is only one form, and it is already moving.
I've been trying more AI products lately with less interest in which model is better or which one gives the best answer.
I'm interested in the interaction around the model.
Chat made AI approachable. Type what you want, get something back, continue the conversation.
But some of the more interesting products I've tried lately are moving beyond that pattern entirely.
And it's changing what I pay attention to as a designer.
Conversation is great at moving forward.
It's less good at holding things still.
Try comparing four options across a long thread. Or editing one part of an output without disturbing everything else. Or remembering which recommendation came from which assumption ten messages ago.
Eventually, products start rebuilding interface inside the conversation.
Tables. Cards. Previews. Structured outputs.
Not a stack ranking. Each one is a form I wanted to feel, not just read about.
Most AI products still communicate through prose. Ask something. Get a paragraph back.
Jev points toward something different: structured decisions instead of just language.
A paragraph asks the user to interpret it. Structured output can become part of a workflow.
The shift isn't better writing. It's moving from answers toward usable objects.
With Instinct, what stood out wasn't the output. It was the channel.
I reached the AI through messages and calls instead of opening a separate product.
A message thread implies continuity. You can leave, come back, receive something, ignore it, respond later.
The channel isn't just where the AI appears. It changes what the AI feels like.
Higgsfield made me notice something else. The work moved onto a canvas.
A generated image wasn't simply the output anymore. It had parts. Elements could move, relight, and change.
The interesting moment wasn't generation. It was what happened after generation.
A generated output can impress me. An output I can manipulate becomes a tool.
When we explored Surfaces at Nayya, pairing conversation with a structured panel still felt relatively new.
The idea was simple: the conversation could move. The important information shouldn't have to.
A transcript can handle questions and explanations. But costs, tradeoffs, evidence, and decisions benefit from somewhere stable to live.
Now that pattern feels increasingly familiar.
Previews. Documents. Canvases. Artifacts. Persistent work beside the conversation.
Designers already collect patterns. Onboarding. Navigation. Checkout. Search.
My references used to be mostly visual.
Now I'm saving behaviors.
Screenshots don't always capture those things.
Sometimes the pattern isn't what the interface looks like.
It's what the product decided to do.
I'm less interested in keeping up with AI terminology than I am in experiencing whether the interaction actually changed.
That usually means using the product. Not just watching the demo.
That's often where the more interesting pattern shows up.
For now, my AI pattern library is deliberately unfinished. It probably should be.
Because the patterns are still evolving.
And if the products are evolving this quickly, our understanding of how to design them has to evolve with them.