Preparing for a post-web world: when AI becomes the interface

For the last thirty years or so, we have designed digital services around a fairly stable assumption: people sit in front of screens, browse websites, fill in forms and click buttons. The web has been an extraordinarily successful interface, and one that higher and further education has rightly invested in.

But that assumption is starting to break down.

We are moving towards a post-web‑ world. Not one where websites suddenly disappear, but one where they fade into the background. Increasingly, the primary interface is going to an AI agent acting on our behalf.

Now is the time to start thinking seriously about how our technology, systems and governance need to adapt.

From interaction to delegation

The terminology around AI agents is often unclear and sometimes used more for marketing than clarity. Instead of responding to a single prompt, these systems can be given a goal and work out how to achieve it. They observe a situation, plan actions, use tools or systems, assess the outcome, adjust their approach and repeat until the goal is met.

As a result, websites stop being destinations and become infrastructure. This change is being accelerated by emerging standards that allow systems to describe what they can do, enabling agents to discover and invoke those capabilities dynamically. As users, we move from interaction to delegation.

This brings a set of challenges for universities and colleges that we need to start addressing now.

Whose agent?

We are likely entering an “bring your own agent” era, where students and staff arrive with AI systems that have followed them throughout their education, carrying deep personal context and preferences. They will expect to use these in an education setting.

At the same time, institutions are likely to provide their own agents. Are we going to expect people to interact directly with institution-provided agents, or are we moving towards agent‑-to‑-‑agent interaction? The latter seems more likely, which means we need to start thinking now about how this will work in practice, both from a technical and governance perspective.

It also impacts our service design. If we design our services only for humans, and actively try to block artificial intelligence systems, we risk making our services increasingly irrelevant.

Designing for humans and machines means thinking beyond websites. It means focusing on actions and capabilities rather than pages and forms and being explicit about how agents are allowed to interact with institutional services.

Security in an agent-mediated environment

We spent years training staff not to click suspicious links. Agents will potentially click them automatically. Prompt injection is social engineering for machines and agent-based‑ systems can act at machine speed using delegated credentials.

This changes how you manage risk.

Audit and logging approaches designed around human users struggle when actions are taken by artificial intelligence across multiple systems. When something goes wrong, we need to know not just what happened, but which system acted, under whose authority, and within what constraints.

Skills, judgement and human responsibility

As autonomy increases, human responsibility doesn’t disappear. We need to understand how responsibility works when we have defined goals rather than precise actions.

Defining goals clearly, supervising agents, exercising critical judgement over outputs, and knowing when not to delegate become essential skills. This applies across professional services, learning and teaching, and research practice.

An agent-mediated‑ future raises many new governance questions. What actions is an agent allowed to take without human confirmation, and where should explicit approval be required? Should an AI system ever be permitted to log in or act as a user? How should environmental considerations be weighed against potential efficiency or productivity gains from more intensive AI use?

These and many other questions need to be addressed soon. We still have a little time to get the answers right, but that window is closing.

What institutions should do now

At a minimum, leadership teams need to build shared understanding of what delegation to AI actually means in practice. Institutions should map which systems are already usable by agents and where vendor roadmaps are heading. Clear principles for delegation, oversight and accountability need to be established. And we need to start designing beyond the web, thinking in terms of actions and capabilities rather than pages and forms.

The web has had a good innings. It won’t disappear, but it will almost certainly no longer be the central focus for users. Now is the time to start planning for this.

To continue the conversation, join our AI community.