What if the website is already obsolete?

my colleague Steven and I were discussing web development when we suddenly found ourselves questioning the whole premise.

What if the website was already obsolete?

It was one of those moments when a familiar subject abruptly opens into something much bigger. We were talking about developing websites, while the way people access information could be changing underneath us.

If an AI agent can find information, interpret it and return exactly what someone needs, how often will that person need to visit the website?

And if the agent can also compare options, check eligibility or complete a transaction, what are we actually building our digital services around?

I find that possibility enormously exciting. It should also bring some urgency to conversations about our next website investment.

Michael Webb, Director of AI at Jisc, captures the shift in his article Preparing for a post-web world: when AI becomes the interface:

“As a result, websites stop being destinations and become infrastructure.”

His argument is that agents will increasingly mediate our interactions with information and services. The pages may remain, while our need to navigate them diminishes. (jisc.ac.uk)

That changes the brief for anyone who owns a website.

For years, we have invested in helping people find their way through our information: menus, landing pages, search boxes, carefully arranged journeys. There is considerable skill in making those experiences work.

But the person arriving with an agent may never encounter any of it. They might simply ask: “Does this apply to me, what will it cost, and what do I need to do next?

Our information needs to be ready to answer.

That means beginning the transition towards discoverable, structured data: information published with clear meaning, consistent relationships and enough context for another system to use it reliably.

Consider a service described across several pages and a downloadable document. A human might eventually piece together the eligibility criteria, charges, deadlines and exceptions. An agent faces the same ambiguity, with the added risk of confidently joining the wrong pieces.

We can make that easier. Publish the criteria explicitly. Identify which charge applies to which service. State when the information takes effect, who maintains it and where the authoritative version lives. Make the relevant data available through appropriate machine-readable formats and documented interfaces.

These foundations already exist in established web practice. W3C guidance covers discoverability, metadata, provenance, standard formats and access to data. AI gives us a compelling new reason to put them to work. (w3.org)

The benefit that excites me most, though, is accessibility.

Every website asks users to learn something about its owner: its terminology, its structure, its assumptions about how people read and find things. Visit ten organisations and you may have to learn ten different ways of doing essentially the same task.

An agent or app chosen by the user could carry their preferences across all ten.

With their permission, it could know that they prefer spoken explanations, need larger text, understand information better through diagrams, or want a short overview before exploring the detail.

The same authoritative information could become a spoken conversation for one person, a visual sequence for another, and a detailed comparison table for someone else. A user could ask follow-up questions in their own words, without having to guess which menu contains the answer.

That offers the prospect of a more personal and responsive form of access. The organisation publishes dependable information; the user’s chosen tool helps present it in a form they can use.

That promise needs careful delivery. An agent must preserve important conditions, distinguish facts from interpretation and let people check the source. Accessible websites and direct routes to services will still matter. But we have an opportunity to extend accessibility beyond the formats each organisation can anticipate and design individually.

We should be eager to explore it.

There are wider benefits too. A maintained source of structured information could support a website, an app, a voice assistant and other services. Clear ownership and versioning could help reduce contradictory answers. Better descriptions of services could make comparison and discovery easier.

None of this requires waiting for a perfect future standard or commissioning a vast replacement programme.

Start with one service and the questions people most often ask about it. Identify the authoritative answers. Resolve inconsistencies. Publish clear, accessible content alongside useful structured data. Give it stable addresses, dates and ownership. Then test whether different tools can find it and explain it accurately.

Structured data alone will not guarantee visibility in AI answers. Google explicitly says its AI search features require no special schema. The work is broader: making information discoverable, understandable, current and trustworthy. (developers.google.com)

The urgency comes from the time that work takes. Untangling information, agreeing definitions and establishing ownership rarely happen overnight. Organisations starting now can learn through small, practical changes while their existing services continue to operate.

Those waiting for falling visitor numbers may discover that their audience has already changed its habits.

I keep returning to that conversation with Steven. We began by discussing web development and ended up wondering whether the thing we were developing was still the centre of the experience.

Our next user may arrive through a tool that already understands how they communicate and what they need.

I want our information ready to meet them.