Two Agents, One Acronym: What Sitecore Really Bought with Scrunch

On June 3rd, Sitecore announced it had acquired Scrunch. Bloomberg put the price at around $225 million, Sitecore itself did not disclose terms. Chris Andrew stays on and Scrunch continues as a division. The trade press filed it under one headline within the day: Sitecore bought an answer engine optimization tool. That framing is accurate, and it is the least interesting thing about the deal.

The interesting part is the category Scrunch says it belongs to. Its flagship product is the AXP, the Agent Experience Platform. Not Agentic, Agent, that one word is carrying more weight than most of the coverage noticed, because right now the industry is using the same acronym for two products that have almost nothing to do with each other.

What Scrunch actually is, at the infrastructure level

Most coverage described what Scrunch does for a marketer. Fewer people looked at where it sits, and this is probably the whole argument.

AXP is not a content workflow, an editor plugin, or a place your team logs in to write. It is middleware deployed at or near the CDN edge. It inspects inbound requests, detects when the caller is a known AI retrieval bot (ChatGPT, Perplexity, Claude, etc), and serves that caller a stripped-down, token-light HTML version of the page it asked for. Everyone else, every human, gets your normal site, unchanged. Same URL, two representations, chosen by who is asking.

If you're an architect, that detail should make you pause. Serving different bytes to a bot than to a human, at the same URL, has a name, and the name is cloaking, and search engines have penalized it for twenty years. Scrunch's answer to that objection is specific. AXP only targets AI retrieval bots, and it explicitly leaves Google and Bing's index crawlers alone, so your traditional SEO surface is untouched. The parallel representation is for the machines that read on a buyer's behalf in real time, not the ones that build a search index.

That distinction is the entire governance case, and I would not wave it through unexamined. It is defensible today because the AI platforms want the clean version and are not treating differentiated delivery as an offense. That is a norm not a guarantee. If a major model provider ever decides that a token-light machine-only representation is a manipulation to be discounted, every brand that built on this pattern inherits that decision at once. Worth knowing what you are standing on before you build on it.

The definition problem nobody wants to name

Here is the confusion. Two completely different things are both being sold as "the agentic experience platform," and the vendors have not agreed on which one owns the term.

One is the agent that visits you. It reads your content on a buyer's behalf, and the job is to make sure it reads you accurately and cites you. That is Scrunch. That is what "Agent Experience Platform" means in Sitecore's press release: the experience delivered to the visiting agent.

The other is the agent that works for you. It sits inside the marketing stack and produces content, builds campaigns, and optimizes journeys. That is what Salesforce means by an agentic marketing platform, and what Adobe means by an enterprise coworker. The experience there is the operator's, or the human's, produced by agents.

Same acronym, but opposite architecture. One optimizes the content an external agent consumes. The other optimizes the work an internal agent performs. The "X" for experience is contested. So is the "A", and not quietly.

You do not have to take my word for the naming being contested. Two vendors shipped a product abbreviated AXP within a week of each other in June, and they do not agree on what the first letter stands for. Scrunch's, now Sitecore's, is the Agent Experience Platform, the edge layer that serves the visiting agent. Contentstack's, announced June 9 and billed as the industry's first AXP, is the Agentic Experience Platform, an operator stack that pairs an agent layer meant to act on your behalf (Agent OS, which it calls its system of action) with content as the system of record and customer data as the system of context, so agents can do the content work while marketers stay on strategy. Same three letters. One is where the agent reads you. The other is where the agent works for you.

Two-column diagram contrasting two kinds of AI agent. On the left, agent-as-visitor: tools like ChatGPT and Perplexity that read your content on a buyer's behalf, served by Scrunch's AXP, the Agent Experience Platform (Sitecore plus Scrunch, at the CDN edge). On the right, agent-as-operator: autonomous content and campaign agents that produce your content, represented by Contentstack's AXP, the Agentic Experience Platform, with Agent OS at general availability, alongside Salesforce with Contentful and Adobe. A bar beneath both columns reads: same three letters, AXP, claimed by both sides for opposite things. — The same acronym, AXP, now ships from two vendors for opposite jobs. Scrunch (now Sitecore) optimizes the content a visiting agent reads. Contentstack optimizes the work an operating agent performs. Deciding which one you mean is the first architectural question, not a labeling detail.

This is not pedantry. If you are an architect being asked to "add an agentic experience layer," the first useful question is which of these two you mean, because they touch different parts of your stack, carry different risks, and are bought from different vendors.

Everyone moved at once

The strongest evidence that this is a genuine category race, and not one company's bet, is the calendar. Within a few weeks of each other, every major DXP planted a flag, and they split cleanly across the two definitions above.

On the visitor side with Sitecore sits Optimizely, which launched a full AEO platform in partnership with Conductor, pairing AI-traffic analytics with prebuilt agents that find and close content gaps. Adobe put generative engine optimization at the center of its story with LLM Optimizer, running a documented analyze, plan, act, adapt loop and tracking brand visibility across the answer engines. These are all answers to the same question Scrunch answers: where are we in the AI's answer, and how do we fix it.

On the operator side, Salesforce put a definitive agreement in place for Contentful and rolled out Agentforce marketing agents that generate omnichannel content from a brief. Contentstack planted the same flag and, on naming, went furthest of anyone, calling itself the first Agentic Experience Platform. Then two caveats I would insist on before repeating any of this as fact. Maturity is uneven: Contentstack took its Agent OS layer to general availability with the launch, while Salesforce's Agentforce Content Agent, the piece that matters most here, was still in pilot as of mid-2026, and the Contentful deal is a definitive agreement, not a closed acquisition. The same discipline applies to Adobe's SEMrush move, which was announced intent, not a shipped integration. Read the category by what has shipped, not by what has been named.

The pattern underneath all of it is simpler than the product names suggest. The DXP vendors are chasing the bots, or more precisely, selling you the technology that helps the bots find you. That is the market Sitecore just paid nine figures to enter directly rather than build.

What this means if you architect on SitecoreAI

The moment you own both a DXP and an edge agent-serving layer, a redundancy question appears that none of the press coverage asked. It is the question that decides whether this acquisition is a feature buy or an architecture buy.

Start with an uncomfortable observation. You already have a clean, structured, machine-readable representation of your content. It is called Experience Edge. Your Content SDK head queries it over GraphQL, gets structured content back, and renders pages. So why would you need a third party to reconstruct a machine-readable representation by re-reading your rendered HTML at the edge?

That leaves an obvious question, Experience Edge and AXP appear to solve the same problem, but they don't. Pull them apart and you find you are running not two representations of your content, but three.

Vertical flow diagram titled "One source, three representations." A single source of truth, the Sitecore content graph in Experience Edge, is queried by the site's own renderer to produce rendered HTML for humans, which is then re-read and simplified at the edge into token-light AXP HTML for AI agents. A dashed arrow runs from the source directly to the agent HTML, marking the cleaner design: project the agent view from the source instead of scraping it back out of the rendered page. — Experience Edge, the human page, and the agent page are three views of the same content. Today the agent view is scraped from the rendered page, a projection off a projection. Owning both ends is what lets you project it straight from the source instead.

Experience Edge is a delivery API. It is authenticated, it speaks your schema, and only your rendering layer talks to it. An external retrieval bot never touches it. ChatGPT does not hold your context ID and does not know your GraphQL schema, it crawls your public URL and eats whatever HTML comes back. So Experience Edge, for all its structure, does nothing for the visiting agent. It was never in that conversation.

AXP lives one layer further out. It sits at the CDN, detects the bot, and hands it a light HTML variant instead of the full human page. And this is where the redundancy is conceptual, not functional.

Conceptually, part of what AXP sells is a patch for a decision you could have made upstream. Scrunch exists because sites are JavaScript-heavy and agents struggle to parse them. A well-built, server-rendered SitecoreAI site with clean semantic HTML and proper structured data is already largely legible to an agent. For that class of site, a slice of AXP's value is covered at the source, by rendering discipline you control, before any edge middleware enters the picture.

Functionally, though, AXP does three things Experience Edge simply does not. It detects agent traffic and separates it from human traffic. It transforms content into a token-light form tuned for a model's context window rather than a browser. And it gives you observability: which agents pulled which URLs, and how that maps to citations and, eventually, revenue. Experience Edge is a cache, not an agent-serving layer and not an analytics one. On that axis the two are complementary, not competing.

So the honest read is that the redundancy is in the representation, not the function. And that is exactly what makes the acquisition architecturally interesting rather than just commercially sensible. Look at the three-representation diagram again and notice the ugliness in it: the agent view is derived from the human view, which is itself derived from the content graph. You are building a projection off a projection. Cacheability, I have argued before, is a property of composition and not something you declare, the same logic applies to representation. The agent-facing view should be projected from the source of truth, not scraped back out of a rendered page two layers downstream.

Today AXP does the scrape, because Scrunch was an external vendor with no access to your content graph. It had no choice but to work from the outside in. Sitecore now owns both ends of that pipeline. The clean design, the one this acquisition makes possible for the first time, is to generate the agent representation directly from the same content graph that already feeds the human one. One source, two projections, siblings instead of a chain.

Whether Sitecore rewires it that way is roadmap, not documented, and I would flag it as the open architectural question rather than assert it as a plan. But it is the strongest version of the strategic case, and it is worth stating precisely because none of the coverage did. This deal is not "we bought a tool that helps you show up in ChatGPT." The defensible version is "we can now generate the machine-facing representation of your content from the same source of truth that produces the human one, inside the governance you already run." That is a platform argument, and it is the only one that justifies the price.

One consequence follows immediately, and it is the practical thing to carry into a discovery conversation. You now have two delivery surfaces to govern, not one. Brand integrity, accuracy, redaction, and freshness used to be questions about the page a human sees. They are now also questions about the representation an agent reads, and those two can drift. A fact you fix on the human page does not fix itself on the agent page unless the two are projected from the same source. If they are not, you have doubled your governance surface and split your source of truth, which is the failure mode this entire category is supposed to prevent.

The bottom line

Strip away the acronym fight and two things are true at once. The shift is real: buyers are forming opinions inside an AI answer before they ever reach your site, and a brand that is invisible or misrepresented there is losing at the top of the funnel in a way its analytics will not even show. And the category built to address it is unsettled, with vendors selling opposite architectures under one name and shipping a fair amount of it as pilot rather than product.

For a Sitecore architect, the move is not to chase the label. It is to ask the two questions the label hides. Which agent are we serving, the one that reads us or the one that works for us, because they are different projects. And where does the machine-facing representation come from, a projection of our source of truth or a scrape of our rendered page, because that answer decides whether we are running one governed content layer or two that will drift.

My interpretation is that Sitecore paid roughly $225 million because controlling the agent-facing representation is strategically more valuable than simply owning an AEO product. Whether it builds it the right way is the thing worth watching, and the thing I will be testing against real tenants rather than press releases.

What's next

The piece I want to write after this one is the practical counterpart: what "agent-ready rendering" looks like on a SitecoreAI head before any edge layer is involved, because a meaningful share of AXP's value is server-rendering discipline and structured data you already control. That connects straight back to the component library work and to the unified data layer, because the same profile and the same content graph that feed personalization are what a well-designed agent representation should project from.

If you are already running an AEO or agent-serving layer in front of a SitecoreAI site, especially if you have measured whether the agent-facing representation drifts from the human one, I would be interested to compare notes.

How we're approaching this at Altudo

This isn't just a theoretical discussion for us. At Altudo, we work with enterprise teams building and modernizing SitecoreAI solutions, and agent readiness has become a recurring part of those conversations. In practice, that means understanding where a brand appears in AI-generated answers, making SitecoreAI experiences structurally legible through server-side rendering and well-modeled content before introducing additional edge layers, integrating discoverability and generative engine optimization into the broader content and personalization strategy, and ensuring that both human and machine facing representations remain governed from a single source of truth.

If your team is evaluating an AEO or agent-serving layer on top of Sitecore, or simply wants to separate product marketing from architectural reality, I'd be happy to discuss the trade-offs. Feel free to connect with me on LinkedIn, or learn more about our work at Altudo.