How AI agents will change the role of your website

Picture the scene. It’s 2035. Your morning coffee arrives through your office window by drone, and you get a notification from your AI agent telling you it’s just paid the final balance on your holiday next month. You reply “Great! Can you check if there’s anything else I need to do?” Turns out your passport is out of date. No big deal – your agent is on it. While you get on with your day, it completes the ‘Apply for a Passport’ service for you, and asks your permission to search your phone to find the right photograph to use. A little later on, it comes back – “You’re all set. Your new passport will be delivered to your mobile phone wallet in 10 working days. Would you like me to complete your tax return next?” You say yes. 

And it all gets done without you ever laying eyes on GOV.UK. 

Of course, AI agents won’t completely replace websites. But websites might not always be the primary way we experience services.

Who is your future website for?

Websites may remain essential in the future, but not only because people need to visit them to get information or complete tasks.

Traditionally, we’ve designed products and services knowing that people will navigate through them themselves. We organise information into pages, create navigation systems, map user needs and design journeys that guide people through a process.

If that process is eventually navigated by an agent rather than a human, some of our work will change. Instead of thinking about how a user navigates the service, we’ll have to make sure an agent can understand it well enough to act on that person’s behalf. That may mean continuing to design well-structured content, but it could also mean giving agents more direct access to the service’s information, rules and capabilities.

Organisations will still need a public account of what they and their services do, and how they work. Regulators will still need to see that information. Auditors may need to trace a decision back to an authoritative source. Journalists, campaigners and researchers may need to examine the rules. People will still need somewhere to go when an agent gives them an answer they don't trust, or when they want to understand why a decision was made. So a website might start to look like a record of the service, rather than the entry point to it.

That creates a different set of expectations. People may still need to use your website, particularly those who can’t or don’t want to access the service through an agent. But it will increasingly need to be a place that supports scrutiny and provides accountability. It will need to make clear what the organisation says, which information is authoritative and what rules and policies an automated system, or any technology acting on a users’ behalf, should be relying on.

Content as infrastructure

Information architecture will have to evolve, too. As well as helping people find the right page, we will need to help machines retrieve the right information, understand when it applies and distinguish it from similar or contradictory guidance.

In other words, content becomes infrastructure, functioning a bit like an API. It gives another system with access to the knowledge and rules of a service without requiring that system to navigate the interface designed for a person – which means content needs to work outside the page on which it was originally published. Interestingly, GOV.UK already has a content API – to allow applications that use content from GOV.UK to keep that content up to date, and there are already third party Model Context Protocol (MCP) servers for GOV.UK content available. These MCPs can act as an overlay on top of APIs to help agents navigate services more easily. One future use case for these could be giving AI agents and chatbots structured access to GOV.UK content as a knowledge source.

But the information an agent uses is only part of the problem. We also need to think about the instructions that determine what it does with that information.

What are we designing now?

If services evolve so that the website is no longer a central part of the user experience, design problems don’t disappear. We’ll still need to test ideas and assumptions. Service rules, handoffs, risk and touchpoints will remain – and continue to be service design problems. 

But content designers, and interaction designers, may find ourselves doing different tasks and working on different parts of the service than we do now. 

We might be designing information for people and machines at the same time. We might be helping to structure the rules and knowledge an agent relies on – thinking about things like semantic structure, and machine–readability. And we might be working on the language that helps determine how an agent behaves.

That starts to blur some of the boundaries between disciplines. And part of our job now is to work out what skills we’ll need to develop — or bring together — to design the services of tomorrow. The recent job posting for the first ever permanent Conversation Designer role at GDS, titled Senior Interaction Designer (Conversation), suggests our responsibilities are already beginning to overlap. At TPXimpact we’ve started to respond to this by working with an AI Conversation Designer to develop training for our content design team, focused on the kind of skills we might need.

As we think about designing the services of tomorrow,  it’s no longer as simple as designing the content, and designing the interactions, because the two increasingly shape one another. To me, the most interesting question isn’t whether websites will still exist. It’s what else we’ll be designing when people no longer need to use them.