Blog

We taught computers to look like our world. Now we’re teaching them to handle the work so we can leave the screens and get back to living.

In 2024, I predicted that software interfaces as we know them would start to go away: the screens, the menus, the workflows that make us first learn how an application thinks before we can get anything done. My claim was that interfaces won’t completely vanish, but their complexity will shrink into a simple conversation. And now, 2 years later. I’m doubling down.

My reasoning comes from a pattern in interface history: we make technology easier by connecting it to something people already understand. First that meant familiar objects. Now it can mean a familiar way to ask for help.

1. We built software out of familiar things

Desktop. Folders. Trash. Shopping carts. Calendars. These metaphors let people bring what they already knew into an unfamiliar place. It’s a lens, not a law, since some conventions, like hyperlinks and swipe, were simply learned. But the pattern is strong.

There is another familiar way we get things done: we ask someone to help. We explain what we want, set constraints, and let them handle it.

A simple conversation.

2. The first real signals

Grocery shopping shows the idea. On a screen, we still walk the aisles: plan meals, check the pantry, compare, swap, trim to budget. Now imagine saying, “Plan five easy dinners for two. Keep it under $100.” The assistant replies with a set of meals and a $94 cart, and asks if that sounds good.

I describe the job. The software coordinates the work.

This is no longer hypothetical. In late 2025, Instacart launched a full grocery app inside ChatGPT, where people can plan meals, build a cart, and check out in one conversation (eWeek). Jakob Nielsen called this shift “intent-based outcome specification,” the first new UI paradigm in 60 years, back in 2023 (NN/g).

This isn’t simply Siri or Alexa again. Those assistants matched commands to a fixed list of skills. Today’s models handle open-ended requests and multi-step tasks far more flexibly, which is what makes delegation plausible.

3. It isn’t just a consumer story

The same pattern fits enterprise work. Picture a finance lead saying, “Approve expense reports under $200 with receipts, and flag anything from a new vendor.” Today that is a queue, filters, and clicks. In a delegated model, the system handles the routine and brings back only the exceptions, with an audit trail the lead can open on demand.

4. The hard problems still require design

Conversation alone won’t deliver this. People struggle to put needs into words, which Nielsen calls the articulation barrier (UX Tigers). Clicking is often faster than typing. Users need to verify what an agent did, and someone is accountable when it gets it wrong.

My UI prediction is a TL;DR interface with depth-on-demand.

Remember “Show and Tell” from elementary school? The future of UI may become as simple as that. An interface that you can tell what task to perform and it can show you results you need to confidently understand and trust that the task is complete.

Tell me what happened, what needs my attention, and what comes next. Then let me go deeper into products, charts, and costs when I choose. The software should interrupt only for real exceptions, and I should control what counts as one. Nielsen expects hybrid interfaces that keep many GUI elements, which fits my view that browsing and direct manipulation do still have a place.

5. What this means for your roadmap

If assistants become the front door, products risk becoming back-ends behind someone else’s interface.

Instacart describes its strategy as reaching shoppers through its own app, retailer sites, partner platforms, or AI agents (The Paypers). Three things to act on:

• Expose your capabilities so an assistant can act safely, with clear permissions, limits, and an audit trail.
• Design the exception path: what the system handles alone, what it escalates, and how.
• Measure outcomes completed, not time spent on screen.

We used familiar objects to help people understand unfamiliar machines. Now we can build around another familiar behavior: asking for what we need. The future I’m betting on is one where I can close my laptop and trust the software to come back when it has a result or needs my judgment.

So which workflow in your product could a customer hand off in one sentence today?

Photo by Christin Hume on Unsplash.