Chat is not the problem. Useless chat is.
There is a lot of talk right now about moving beyond chatbots as AI interfaces. You see people saying things like, “Don't just tack a chatbot on top of a 2010 UI and call that innovation.” And sure, I get the point. A useless chatbot sitting in the corner of an app, answering support questions badly, is not the future of software.
But I also think a lot of the people saying this are missing what is actually happening.
If you spend enough time with AI tools, especially as a builder, you start to notice something obvious: we are being trained to communicate with software through natural language. Text, voice, video, whatever the format is, the behavior is the same. We tell the machine what we want, and we expect it to figure out the steps. That changes everything. Once you get used to giving software instructions directly, clicking around to figure out where everything lives starts to feel ridiculous.
As a builder, I already feel this very strongly. I now have almost no interest in updating an app or website that is not plugged into Codex, Claude Code, or some kind of coding agent. If I have an old project built with a third-party tool, and I cannot easily connect it to an agent, I would rather rebuild the whole thing from scratch than spend hours clicking around, reading docs, untangling somebody else's structure, or trying to remember how the thing was put together.
I do not want to do that anymore.
I want to give commands.
That may sound extreme, but I do not think it is just a personal preference. I think builders are early indicators of where regular user expectations are going. Once people get used to telling software what they want, they become much less patient with products that force them to learn the product's internal logic first.
And this is the part I think people are underestimating.
I don't think people hate chat. I think they hate chat boxes that sit there pretending to be useful while the user still has to do all the work.
A chatbot that can only answer FAQs is not interesting. A chatbot that says “I can help with that” and then links you to a help article is not interesting. A chatbot that makes the user do all the work anyway is not innovation.
But a chat interface that can actually operate the product? That is different.
If I can say, “Import these transactions, categorize them, find the duplicates, and show me what changed from last month,” that is not a gimmick. If I can say, “Create a budget based on my last three months of spending, but be stricter on eating out,” that is not just a chatbot. If I can say, “Find the subscriptions I forgot about,” or “Show me why my balance is lower than expected,” or “Prepare a clean tax summary,” that is a new control layer for the product. That is the future I care about.
Not chat for the sake of chat. Not AI sprinkled on top of the UI so the company can say it has AI. But natural language as the fastest way to get outcomes from software.
Because the truth is, most users do not want to learn your app. They do not want to understand your navigation, your filters, your settings, your terminology, your workflows, or your clever new dashboard layout. They want the result. They want the thing done.
The old software model made users adapt to the product. The new software model should make the product adapt to the user.
That does not mean UI is dead. People still need buttons, screens, charts, tables, confirmations, and visual feedback. Especially in products where trust matters. If an AI creates a budget, I still want to review it. If it categorizes transactions, I still want to see what it changed. If it gives me financial insight, I still want the receipts.
So no, the winning interface is probably not “chat instead of UI.”
It is chat as the command layer, and UI as the place where users verify, edit, understand, and trust what happened.
That is why I think the anti-chatbot take is too simplistic. Yes, do not tack on a useless chatbot and call it innovation. But if the chatbot can help users control the product, create new things, skip unnecessary steps, and get to the outcome faster, then yes, absolutely tack it on. Put it right on top of your 2010 UI if you have to.
Because at product-market fit, users are not grading your interface like designers on Twitter. They are asking one question:
Can this thing do what I need?
And increasingly, the easiest way for them to ask for what they need is just to say it.