The website was never the point: how agentic AI is changing digital experience
The website used to be the product. Now it's one rendering of something deeper. Lorna Foott on the four-layer shift changing digital experience platforms and what organisations need to get right.
Lorna Foott, Director of Partnerships, 28 July 2026
For nearly thirty years we sold websites. That's how it looked, anyway.
What we were actually selling was a way for an organisation to manage what it knew and get it in front of the customer who needed it. The website was the visible edge. The CMS was the machinery behind it, and for most of that time the machinery was the thing you bought, compared and fought over in procurement.
That era is ending, or rather evolving, and faster than most of the market wants to admit.
What's happening across the DXP category
Look at the last few months across the digital experience platform category. Salesforce agreed to acquire Contentful. Sitecore bought Scrunch, an answer-engine optimisation platform. Optimizely partnered with Conductor and rebranded around agentic AI. Adobe renamed Experience Cloud as CX Enterprise and reorganised its whole customer experience business around an agent-orchestration layer.
Four companies, four moves, and one argument underneath all of them. At the sharp end of the enterprise market, the platform's no longer the only thing organisations are buying. The agent layer is the headline act, and the content system that used to be the product becomes something quieter and more important: the substrate that agents draw on.
An agent works across four layers: the substrate it draws on, the retrieval that finds it, the orchestration that acts on it, and the experience a customer meets. All four are changing at once.
The maturity conversation has shifted
A year ago, agents that could reason across a customer conversation and act with the right context were demo territory. Now they're in production.
We've seen it. Working with Optimizely, we built a seven-agent AI ecosystem for Corinthia Hotels on Opal's agentic infrastructure, personalising the experience across five distinct guest personas and a fourteen-stage lifecycle, while protecting Corinthia's brand voice at every step.
10x productivity gains in variant production.
100% first-time approval rate, with zero brand violations.
Zero hallucinations across all outputs.
150+ person hours saved at strategic inception.
The work has since won two awards: Artificial Intelligence in Travel Marketing at The Travel Marketing Awards, and Innovative Use of AI in Marketing at the Hotel Marketing Association Awards 2026. The judges called it a rare example of genuine innovation in an industry not known for moving fast. Dynamic, AI-personalised experience at scale, holding to brand standards a human would recognise, is no longer a promise. It's running. Read the full case study.
Features have become a canvas
A feature used to be a fixed thing on a spec sheet, something you ticked off against a competitor. Now it's a starting point. The underlying capabilities of an agent-capable platform can be assembled in combinations no vendor specified in advance, and no two organisations need build the same thing from the same foundation. The Corinthia ecosystem is one shape that took. There will be others nobody has built yet. The ceiling hasn't moved so much as disappeared.
Content strategy has moved to the boardroom
Content strategy has finally become a boardroom conversation, and not because content teams got louder. It has meant brand, thought leadership, SEO and campaigns, all well-resourced. The parts that agents now depend on, content modelling, structure, governance and taxonomy, have lived inside content operations and been treated as BAU.
The argument used to be "your content should be structured because of best practice." Now it's "every one of the machines reading your brand is making a decision about it, and none of them are guaranteed to pull what you want from a page written for a human on a scroll."
Content's remit has expanded well past the website. Product data, help content, policy documents, internal knowledge, brand governance rules, customer history. All of it is content an agent could potentially draw from, and almost none of it is what a content team traditionally owns.
From website strategy to experience strategy
You're no longer planning a website. You're planning the connection across every surface where it happens, and connecting and structuring what your organisation already knows so that everything downstream, human and machine alike, has something reliable to work from. The website is one rendering of it. The AI answer is another. The agent acting for a customer is a third. Weak substrate underneath, and every rendering is weak. Even the best orchestration works with what it's given.
Most organisations quietly assume an agent can just read what's there. It can, in the sense that it'll take something. It could also be the wrong thing, arrive without context, and give you no way to correct the answer. Structuring content across those systems is the discipline that hands the agent something reliable to work with. Without it, you've handed over control, accuracy and governance in one move.
The imagination gap
Getting the substrate right doesn't just prevent mistakes. It changes what you can imagine building. When it comes to project ambition, two questions pull in different directions. How do we do what we already do, faster? And what could we do that we couldn't before? The first is a stretch on today. The second is a new destination.
There's real value in the first. Automating a known process tightly, with the right controls, is where most enterprises are sensibly starting. But the second question is where the interesting work sits, and it's the one fewer teams are asking. A case review compressed from weeks to hours, where an agent reads the whole file and comes back with a decision. A service function that acts on a problem before a ticket even lands. These aren't horizon scenarios, they're already being built by teams that did the hard work first. And the hard work is unglamorous: clear business analysis, honest data, and different teams learning to work together. Ambition needs those foundations, or it doesn't get anywhere.
The four-layer view
The full picture has four layers, and the website isn't gone from it, just repositioned.
Substrate is the material, distributed across systems.
Retrieval and context make it usable.
Orchestration is what compounds over time, workflows, experiment history, governance, the marketing harness.
Experience is where all of it lands, and the website is one form of that, alongside the AI answer, the agent conversation, the personalisation surface.
Retrieval is standardising. Orchestration is where the durable advantage sits. Experience is where the customer meets any of it, and shaping that is what the whole stack exists to do.
These are exciting new frontiers. The convergence is real, the direction is right, and the platforms leading it are building things worth building. The rest is craft, and imagination.
The website was never the point. Neither is the agent. What the customer meets, the answer, the interaction, the sense that a brand understood them, is the point. Everything else exists to make that possible.
Talking to us
Getting there is deep work: connecting content across systems, structuring it for machines without breaking it for humans, and designing the experience customers actually get. If you're thinking about how the four-layer view applies to your organisation, we'd be glad to talk. Please reach out.
Frequently asked questions
What is an agent-capable DXP?
An agent-capable DXP is a digital experience platform that can build and orchestrate AI agents. These agents can act with some autonomy, reasoning across a multi-step workflow, drawing on structured content, and delivering personalised experiences to customers in real time. The category is evolving quickly, with major vendors including Optimizely, Adobe, Salesforce and Sitecore all repositioning around agentic capability.
What is the four-layer DXP architecture?
The four-layer model describes an agent-capable stack. Substrate is the distributed content and data agents draw on. Retrieval and context make that substrate usable in the moment. Orchestration coordinates what agents do, and compounds value over time. Experience is where customers meet the output, whether that is a website, an AI answer, or an agent conversation.
Why does content substrate matter for AI agents?
An agent is only as good as what it draws on. Without well-structured substrate distributed across systems, an agent will still return an answer, but often the wrong one, without context, and with no way to correct it. Investing in the orchestration layer without doing the substrate work produces expensive systems that agents can't reason over well.
How do we know if we have a substrate problem?
If your organisation has invested in AI capability and still finds agents returning outdated policy documents, missing important product information, or giving different answers to the same question depending on the day, that's a substrate problem. It usually shows up when the front-end AI is working but the outputs feel unreliable. The underlying issue is almost always content: distributed across systems that were never designed for machine consumption, structured for human readers rather than machine parsers, and missing the governance that lets you correct answers at their source. Practical signal: if you can't easily trace an agent's answer back to a specific piece of content that you can then update, you have a substrate problem.
Should we prioritise substrate work or agent work first?
The honest answer is both, but with sequencing. Substrate work takes longer than most enterprises expect and rarely produces visible results in the short term. Agent work delivers visible outcomes faster but is only as good as the substrate underneath. Most organisations that get this right start substrate work in parallel with agent pilots, using the pilot experience to reveal exactly which substrate gaps matter. Substrate work done in isolation, before any agent context, tends to over-engineer for problems that never arrive. Agent work done in isolation, without substrate discipline, produces demos that don't scale into production. The sequencing is start both, let the pilot expose the gaps, then invest deeply in the substrate work the pilots identified.
What should we look for when evaluating an agent-capable DXP?
Four questions to ask any vendor. First: how does your platform work with content distributed across systems we already have, rather than expecting everything to live in your CMS? Second: what does your orchestration layer actually compound over time, and what evidence do you have of that compounding delivering value for existing customers? Third: how do we govern, audit, and correct what agents do when they get something wrong? Fourth: what happens when we want to swap out a component of the stack in three years? The vendors worth taking seriously will have specific answers to each. The ones that talk mainly about individual features are still competing on the old ground.
What is the imagination gap?
The imagination gap is the difference between two questions. How do we do what we already do, faster? And what could we do that we couldn't before? Most enterprises start with the first, automating known processes with the right controls, and there's real value in it. The second is harder and rarer: using agentic capability to build something that didn't exist before, rather than speeding up what already does. A customer service function that resolves an issue before the customer notices it, say, instead of a faster ticket queue. The imagination gap is the distance between those two ambitions, and it's where the work that changes an organisation, rather than just improves it, actually sits.
NEWSLETTER _
Expert Insights
Get expert insights on digital transformation, customer experience, and commercial impact delivered to your inbox.
SIGN UPRelated articles _
