top of page

The Death of UI Design (And the Rise of the AI User)

Writer: Michael Shmilov
Michael Shmilov
1 day ago
6 min read

About ten years ago, when I was hiring designers at Viber, it really bothered me when candidates called themselves "UI/UX Designers."


It wasn’t about the people. I worked with great UI designers who also had great UX instincts. It was that damn slash. Two different crafts squeezed into one fashionable title. Of course we never rejected anyone because of punctuation. We hired based on skills and what the team needed.


But the slash still annoyed me.


But over the next few years things happened anyway. UX wasn’t just for designers anymore. Product managers, developers and entire companies became obsessed with customer satisfaction and delightful experiences. We spent hours obsessing about micro-interactions, pixel perfection and frictionless user journeys.


Today the ground is shifting again and that old debate over the "/" seems almost ancient history.


AI can now spit out interfaces in seconds. Digital products are starting to look more and more similar, predictable, and frankly a little numb.


My reaction? Who cares.


Because something much bigger is happening.


More and more the interface isn’t even being used by a human — it’s being driven around by an AI agent.


UI design is not dying because AI can generate screens. It’s losing its place as the primary way users interact with digital products.


For a human good UX is less cognitive load, intuitive navigation and sometimes visual delight. For an AI agent good UX (or Agent Experience AX) is fewer interactions, faster execution, more reliable results and ultimately what the user asked for.


And that changes how we should think about building products.


What This Means for Product Managers

In the short term, AI feels like a massive shortcut for product teams. PMs, designers and developers can execute faster; prototype ideas; validate concepts at record speed.


But the bigger change isn’t how quickly we build interfaces. It’s who, or what will use them.


We have to start designing for two types of users — humans and the AI agents that work on their behalf.


An agent doesn’t care if you have beautiful CSS, clever onboarding animations or intuitive navigation bar. It only cares whether it can do its job reliably, accurately and efficiently — without wasting tokens, getting stuck in modals or hitting dead ends.


This is not another version of SEO or web performance optimization. Those still matter and discoverability does too but optimizing for agents brings a new goal: autonomous task completion.


If a human visits your website you want to engage them, help them navigate and have a great experience. If an agent visits you want to make the relevant capabilities and information available with as little friction as possible.


And ideally the agent shouldn’t have to navigate your visual interface at all.


The Five Capabilities of an Agent-Ready Product

When I was laying out the 2027 product strategy for [EcoMoat](https://www.ecomoat.ai) I had to translate this shift from high-level idea into product requirements.


How do we make EcoMoat useful not only to investors visiting our website but also to AI agents who research companies on their behalf?


We found five capabilities. We’re applying them to EcoMoat but I think the framework is also applicable to many digital products preparing for an agent driven ecosystem.


1. Discoverability

For an agent to use your service it needs to find you. Traditional search indexing still matters. Many AI assistants use search engines, indexed content and other sources to find information. But being discoverable by agents also means authoritative well structured content that is understood, referenced and cited. If your product isn’t in the information sources that agents use you may never be part of their consideration set.


2. Machine-Readable Intelligence

Stop asking agents to scrape visual layouts, navigate tabs inside tabs or extract numbers from images of charts. Give them structured information (clean APIs, JSON or Markdown where possible). Just as important: Context for the data — timestamps, units of measurement, scoring scales, plain text explanations and direct references to sources. For instance an EcoMoat moat score of 85 means nothing unless you know how the scoring was done, what factors were used and when the analysis was run.


Make your data easy to access, understand and interpret.


3. Change Intelligence

This is one of the capabilities that I’m particularly interested in. Static data alone is not enough for research and decision-support products. Agents need to know what changed, why it changed and whether it matters. If an agent asks for the price of a product don’t just give her the current number. Tell her what happened in the past. Has it gone up? Is it unusually low? What changed recently?


The same applies to investing. If EcoMoat says that a company’s competitive position has changed then an agent should know which factors changed, what new evidence supports the conclusion and how significant the change is. The delta and reasoning are what saves the agent from having to reconstruct the whole story itself. And it makes your service much more useful.


4. Trust and Verifiability

AI agents can make mistakes, interpret information incorrectly or confidently repeat unsupported conclusions. Black-box answers make those problems harder to detect.


This is why we’re investing in traceability at EcoMoat. Conclusions need evidence; analysis dates; methodology versions; clear explanations of uncertainty. If an agent receives an investment analysis it should be able to understand where the conclusions came from, verify supporting information and even challenge the findings.


The goal isn’t just to give an answer. It’s to make that answer trustworthy and verifiable.


5. Reusable Integration

An agent shouldn’t have to “visit” your website like a human tourist every time it needs something. Your core product capabilities should be exposed via well-defined APIs or agent oriented protocols (MCP: Model Context Protocol). That way agents can get information, run authorized workflows and monitor changes programmatically. This also means thinking about permissions, authentication and what actions an agent can perform on behalf of a user.


And why stop at letting agents request information? We could also let them subscribe to updates, through webhooks, email, or other notification channels.


Imagine an agent following a list of companies and automatically receiving updates whenever EcoMoat detects an important change. No need to keep checking.

We might even offer paid subscriptions for AI agents. Well, technically their human owners would probably pay, but I like the idea of having agents as paying customers. :-)


For me this is where a website starts becoming something more interesting: a service that can be plugged directly into a user’s AI workflow. And that brings us to a much bigger strategic question.


The Bigger Question: Who Owns the User?

For years companies have invested heavily in their websites and applications — not only to do functionality but also to build brands, relationships, habits and customer loyalty.


Now imagine an investor who uses EcoMoat’s research, financial analysis and valuation models entirely through ChatGPT or another AI assistant.


The investor gets the value without ever visiting EcoMoat. That’s great for distribution but who owns the relationship with the customer?


EcoMoat provides the intelligence. Another platform provides the experience. Who gets the credit? Who builds the habit? And finally, who captures commercial value?


I think this is one of the biggest strategic challenges ahead for digital products. Being accessible to agents isn’t enough. We need to figure out how to be differentiated, trusted and commercially relevant when someone else controls the interface.


The same question applies to travel platforms, marketplaces, financial services, SaaS products and many other industries.


The Strategic Catch: Agents Have Amnesia

Not literally. AI assistants can remember preferences, context and use connected services.


But here’s the trap I expect many product teams to fall into — assuming that being agent-friendly automatically creates loyalty. Just because ChatGPT, Claude or Gemini got information from your website today doesn’t mean it will automatically prefer your service tomorrow (let alone recommend it to another user).


A good one-off interaction doesn’t make a relationship. That’s why I think the fifth capability, reusable integration is particularly important. Discoverability may get you the first interaction. Repeated usefulness, explicit integrations and user trust are what can make your product part of someone’s regular AI workflow.


Instead of hoping an agent finds you every time, you want users and developers to connect your service deliberately because it delivers value consistently. That is very different from just optimizing a website for traffic.


And who knows? If EcoMoat ends up serving thousands of happy AI agents, maybe we'll introduce an annual "Agent of the Year" award. The winner gets an all-expenses-paid vacation: three days in an AWS data center in Alaska. Plenty of cooling, unlimited electricity (subject to fair usage), and absolutely no humans asking it to summarize another PDF.


The Next Generation of Product Experiences

Human interfaces aren’t going away. And I’m not saying we should stop caring about good design either. But increasingly humans won’t be the only ones using our products.


We’re moving into a world where we have to design for two types of users - people and the AI agents working on their behalf. For product managers that’s a paradigm shift. We need to re-think how our products are discovered, accessed, trusted and ultimately valued. The next generation of great digital products won’t just look good and feel good to use.


They’ll also be easy for AI agents to understand and put to work.


And maybe the biggest challenge won’t be designing the interface anymore but making sure our products still have value when someone else owns it.

 
 
bottom of page