• Most websites and apps still make customers do the work
• AI agents can take that effort away
• Adoption is moving fast and customer expectations keep rising
• Getting this right is a design challenge not a technology one
• Agentic UX Tips: start with friction, get content ready, be open, design for the real moment, set limits, build in accessibility, measure outcomes
• Understanding people is still an important role of the designer
• Build websites that work better for AI assistants

Is your website or app actually valuable to its users?

And by valuable I don’t mean just looking good and ticking a box on someone’s roadmap. I mean really helping people do what they need to do: find the right product, book the best value trip, sort out the bill, get a useful answer.

For most brands, if we’re being honest, it’s less than we’d like. Despite good UX design intentions, for twenty years, organisations have built websites and apps that make the user do the work. Click through the menu, fill in the form, dig out the FAQ, and if that fails, try again over the phone, if there’s a contact number to be found, that is. Much to Steve Krug’s* horror, no doubt, too often we’ve made people think.

AI is changing all that. Rather than designing screens for people to work their way through, we’re now designing systems that take on the cognitive load so people don’t have to.

From AI features to AI agents

First, a quick definition, because “agent” is now getting thrown around about as loosely as “AI” is. Most of what brands have shipped so far is AI as a feature. A chatbot stuck on the contact page, a “summarise this” button, some generated product descriptions. Not the worst, but not exactly groundbreaking.

An AI agent doesn’t just answer questions, it does the work. It figures out what someone needs, takes the steps to get there using your systems and data, and knows when to hand over to a person.

It’s happening quickly, too. Genesys, who work in this space, published their 2026 State of Customer Experience report in July and found that 40% of CX organisations are already using agentic AI, and 82% of CX leaders expect autonomous AI agents to orchestrate the customer experience within three years.

Customer expectations are going up at the same time. In the same research, 92% of consumers said they expect every organisation to deliver an experience on par with the best they’ve ever had. So you’re not just being compared with your competitors anymore. You’re being compared with the best experience your customer has had anywhere… ever!

And the bar is going to keep rising, faster and faster.

Why this matters for UX

Customers aren’t asking for agents. They just want less hassle and an experience they can trust, and some recent research, including from here in Ireland, backs that up.

In August, CCMA Ireland asked over 1,000 Irish consumers how they feel about AI. Most are happy to let it handle the simple stuff, like payments, bookings and tracking a delivery, but when it comes to complaints, billing or fraud, they want a person. 77% want to know when they’re dealing with AI, but only 57% say it’s clear right now.

That’s our design brief right there.

We need to understand what people are actually asking, remember the information they’ve already given and make it easy to reach a real person if needed. These are user experience design problems more than technology ones.

3 examples from our own work

Each one of these examples uses AI to improve the user experience in a different way.

#1 Fáilte Ireland: an AI planner that got more use than site search

We’ve all been there. You fancy a weekend away but have no idea where to start, and a set of filters isn’t going to help you figure it out.

For its Winter in Dublin campaign, we designed and built the AI Short Break Planner for Fáilte Ireland’s Visit Dublin site. People ask questions in plain language to get ideas and information for their trip. It works out what they’re looking for before it retrieves anything, so every answer comes from Visit Dublin’s own content.

Over four months, the planner was used 2.5x more than the site’s search function, achieved a 37% referral conversion rate, and helped visitors articulate intent they couldn’t easily express via filters. We also made it clear to people that they were using an AI tool, which matters given the trust research above.

What I like most about this one is that it doesn’t replace the website. It just gives people a better way in.

#2 Croke Park: Ireland’s first AI stadium assistant

If you’ve ever tried to find the nearest food stall or first aid point with 80,000 people around you, you’ll get why this one matters.

Working with Fexco and Croke Park management, we helped design and launch an AI-powered Stadium Assistant, accessed via a QR code on the back of 5,000 seats. We looked after UX research, landing page design, security, load testing, accessibility and support on the day.

The big UX lesson here was context. People were standing in a stadium, on their phones, in a hurry, so the design had to show them within a few seconds that it was useful and could be trusted. There was no app to download and no sign-up.Over 1,300 attendees used the assistant on launch day, and Croke Park is now planning to roll it out across all 80,000+ seats.

#3 Behind the scenes: agents that protect quality

Not every AI agent needs to be customer-facing. Some of the biggest UX improvements come from agents working inside our delivery process.

Our QA team has evolved AhTestHub into four AI agents covering test generation, automation, accessibility and performance. WCAG 2.2 AA checks are now built into every test suite by default, instead of being left to a separate pass at the end.

These agents aren’t carrying out tasks directly for the user, but users still feel the difference: product improvements get released much faster and, more importantly, when they launch they’re more accessible and less likely to break.

7 things brands should be thinking about now

I know a list like this can oversimplify things. No organisation runs off a checklist. But from what we’re seeing with our clients, these are the things that make the difference between AI that improves the experience and AI that just adds another layer to it.

#1 Start with friction, not features

Start by looking at where your customers are putting in the most effort now. The searching, the comparing, having to explain themselves twice, getting bounced from one channel to another. That’s where an agent can really earn its place.

#2 Get your content and data ready for agents

An agent is only as good as what it has to work with. The Fáilte Ireland planner worked because every answer came from Visit Dublin’s own content. Content that’s structured, up to date and well looked after is now a UX issue, not only a CMS one (I wrote more about this in my AI visibility blog).

#3 Be open about it, and give people a way out

Tell people they’re talking to AI, show them how it got to its answer, and make it easy to reach a person.

#4 Design for the real moment, not the demo

Context changes everything: the device someone’s on, where they are, how much of a rush they’re in. Something that works perfectly in a meeting room demo can fall over in a busy, noisy, real-world setting. Go and see the actual moment people will use it, then design for that.

#5 Decide how much freedom to give it

Not every task should be fully handed over. For each journey, decide whether the agent should inform, recommend or act, and where a person needs to sign off. Start small, show it works, then build from there.

#6 Build in accessibility from the start

If an agent becomes the main way people deal with your brand, it has to work for everyone. Treat WCAG and the European Accessibility Act as design inputs from day one, not checks at the end of the project.

#7 Measure outcomes, not novelty

People using a new feature doesn’t mean it’s working. Measure task completion, conversion, time saved, fewer calls to your support team and satisfaction against the journey it replaced. And be careful about judging success by how many conversations the AI handles without passing them to a person. If a chat ends without a handover but the customer rings you two days later, you haven’t solved anything.

Understanding what people need is still our job

An AI agent is only as good as the thinking behind it. Someone still has to decide what it should do, what it should never do, and what it should feel like to deal with. That’s still our job.

For design teams, it’s a real shift. We’ve spent years designing screens, and now we’re designing how something behaves: what it knows, how it talks to people, when to step in and when to leave them alone. It’s new territory, but the core of it feels very familiar. Understand people, then design around them.

I believe that the brands that get this right won’t be the ones shouting loudest about AI. They’ll be the ones whose customers barely notice it, because things just got easier.

It’s a big part of why we built Flow, our AI-powered delivery model where humans direct and agents deliver.

And it’s not only your own AI you need to think about. More and more, your customers will send their own AI assistants to do things for them, like booking, buying or checking an order. If your website is hard for those agents to read and use, they’ll move on to someone else. With Flow Discover, we help make sure your site is built so AI can find it, understand it, and get things done for your customers.

If you’re wondering where agents could make the biggest difference for your customers, I’d be happy to talk it through. You can book a demo here.

* Steve Krug is a UX consultant and author of Don’t Make Me Think, one of the best-known books on web usability.