Dr.Jingle Intelligence Note

The Website of the Future: Every Visitor Gets a Different Page

Adobe's Carlos Sanchez demos agentic sites at AIEWF 2026: assemble a unique page per visitor in 1–2 seconds from your own content corpus—not hallucinated copy.

Share
X LinkedIn

Key Takeaways

  • At the June 2026 AI Engineer World’s Fair in San Francisco, Adobe Chief Scientist Carlos Sanchez demoed an “agentic site”: the system reads visitor behavior and search terms, infers intent, and assembles a page unique to that person in 1–2 seconds.
  • Sanchez calls this “audience of one.” He told Latent Space: “A lot of people don’t even think you can generate a webpage on the fly… No, you can do it now. This is not the future, this is now.”
  • For twenty years, “personalization” meant picking from preset options—recommend a product, swap a module, change copy. The page skeleton stayed the same.
  • The jump is from “thousand faces, thousand components” to “thousand faces, thousand whole pages”—pages that do not exist until the request arrives.
  • Sites must target both humans and AI agents; WebMCP and related proposals aim to expose structured tools so agents need not scrape the DOM.

One-Line Summary

The internet always wanted the right person to see the right content—it just could not afford “right.” Latency and compute prices are rewriting that arithmetic, from 1999 segment-based personalization to 2026 real-time page assembly.


Body

June 2026, San Francisco. At the AI Engineer World’s Fair (AIEWF), Adobe Chief Scientist Carlos Sanchez stood at the demo stage. He was not showing yet another chatbot, but what he called an “agentic site” web experience: the system reads a visitor’s browsing behavior and search terms, judges intent—browsing, researching, or ready to buy—and then retrieves assets from the brand’s existing content library to assemble a page built for that individual in 1–2 seconds. Sanchez calls this approach “audience of one.” After the event he told Latent Space: “A lot of people don’t even think you can generate a webpage on the fly… No, you can do it now. This is not the future, this is now.”

Demo Video (Project Page Turner)

Adobe Summit 2026 Sneak live demo: real-time page assembly by visitor intent.

1999: The Holy Grail Was Always There—Pages Stayed the Same

Rewind to the dot-com era and “personalization” was already fixed vocabulary in web operations. E-commerce sites recommended products from purchase history; marketing teams placed visitors into audience segments and showed different banners. The Latent Space author recalls managing sites back then: personalization ran for more than twenty years, but it was still choosing from preset options—recommend one product, swap one block, change one copy set. The page skeleton did not move; only a few modules did. That logic differs from Sanchez’s demo. In the demo, a visitor interested in camping opens a coffee-machine brand site and the entire page—copy, product layout, supporting content—is recomposed into an “outdoor coffee” narrative; another visitor types “Europe AI conferences” into an open search box and the system assembles a page around that query on the spot. The page did not exist before the request. That is the leap from “thousand faces, thousand components” to “thousand faces, thousand whole pages.”

Not Hallucination—Assembly From Your Own Content Library

Engineers’ first reflex is often hallucination and fabrication. Sanchez’s answer: the grounding corpus is the brand’s existing content—product specs, blogs, event pages, FAQs, stored in a CMS such as Adobe Experience Manager. Large models interpret intent, retrieve, and orchestrate layout—they do not invent a company that does not exist. Adobe named a related concept Project Page Turner at the 2026 Summit Sneaks segment; the demo site is arco.coffee and technical notes live at of1.live. Third-party technical coverage reported inference on low-latency hardware such as Cerebras; the model can be swapped, but perceived wait for the visitor cannot exceed roughly 1–2 seconds—a completely different path from “render static HTML first, swap modules asynchronously.” Sanchez offered a rough cost estimate of about 1–2 cents per generated page, and added: “That’s just today’s price. Who knows what it will be in six months.” If those numbers hold, marketing pages, landing pages, and long-tail intent pages—scenarios where building a separate site per micro-audience never paid—start to make economic sense. But Adobe has not deployed this at scale on production customer sites; Sanchez said they are looking for brands willing to experiment, with commerce (conversion-focused use cases) as the most natural entry—“any domain that needs higher conversion and has a large matrix of user types.”

Easy to Build, Hard to Know What to Build

The demo is flashy; uncertainty is just as real. Sanchez’s line: “With AI, it’s very easy to build things, but it’s hard to know what to build.” AI makes building easier and choosing harder. They built it first, then went looking for customers. That is not only Adobe’s puzzle. In 2026, site owners are simultaneously testing in-site chat, structured content for agents, generative UI, personal-assistant entry points, and how to pull users back from third-party AI products such as ChatGPT. Sanchez summed it up as “a combination of all kinds of crazy ways”: the user is in chat and you want to show UI; the user is on the site and you want to route them elsewhere; the user is in an OpenAI dialog and you want them back on your property—everyone on the marketing side is feeling their way across the river. The of1.live footer is blunt: this is early-stage technology from the Adobe Experience Manager team. Some Sneaks projects ship; some do not. Do not bet the entire 2026 roadmap on this one demo.

Another Visitor: The AI Agent

If the story stops at “pages humans see,” half the plot is missing. Personal AI assistants increasingly research, compare prices, and reorder on a user’s behalf. When an agent visits, the preference profile it carries is often richer than cookies and recent browsing—it may order toilet paper for you but will not buy a coat without having “seen” it. Sanchez expects sites to target both humans and agents. Will there be “two versions” of a website? He said boundaries may blur, but “obviously you have to target both ends.” On the technical path, proposals such as WebMCP (Web Model Context Protocol) let sites register checkout, filtering, and search as structured tools with JSON Schema via navigator.modelContext, so agents need not scrape the DOM or simulate clicks; backend MCP / A2A can let agents transact with each other without a traditional visual interface. In February 2026, the Google Chrome team released an early preview of WebMCP; the spec is maintained by the W3C Web Machine Learning Community Group and remains a draft community report, not a formal W3C standard. The same page may have a visual layer for humans and a tool layer agents can call—or beyond the human-facing site, a separate agent-to-agent surface. “Agentic targeting” is almost certain to arrive; product form factors have not converged.

Will Websites Disappear?

That was another big question outside AIEWF. Sanchez’s view is conservative: websites in the traditional sense are unlikely to vanish wholesale, but their role will shift. They may cease to be a fixed set of URLs where everyone walks the same navigation path, and become instead a content and interaction system governed by brand rules—assembling the right interface for this visit when needed, and in other scenarios exposing structured capabilities to agents. From 1999 segment-based recommendations, to 2026 real-time page assembly, to agent-readable toolized sites, one thread runs through: the internet always wanted the right person to see the right content—it just could not afford “right.” Latency and compute prices are rewriting that equation. Will websites “assemble themselves”? For some visitors, some intents, some brands—the demo already says now. Whether it becomes mainstream still depends on something older: whether people will trust a machine to choose the layout for this one visit.


Compiled from public sources and commentary. This does not constitute investment advice.

Sources

Conclusion

At AIEWF in June 2026, Adobe’s Carlos Sanchez showed that agentic sites can assemble personalized whole pages in seconds from a brand’s own content—not by swapping preset modules but by generating pages that did not exist until the request. Sites must also serve AI agents via structured tool layers. See sections above for detail.

FAQ

What is this article mainly about?

A: It covers Adobe’s “agentic site” and “audience of one” demo, how it differs from twenty years of segment-based personalization, economics and deployment uncertainty, and the parallel need to target AI agents—not just human visitors.

Demo video (Project Page Turner)—what are the main points?

A: See section “Demo Video (Project Page Turner).” Based on source materials; not investment or legal advice.

1999: The holy grail was always there—what are the main points?

A: See section “1999: The Holy Grail Was Always There—Pages Stayed the Same.” Based on source materials; not investment or legal advice.

Not hallucination—assembly from your content library—what are the main points?

A: See section “Not Hallucination—Assembly From Your Own Content Library.” Based on source materials; not investment or legal advice.

Easy to build, hard to know what to build—what are the main points?

A: See section “Easy to Build, Hard to Know What to Build.” Based on source materials; not investment or legal advice.

Does this article constitute investment advice?

A: No. This is informational commentary and opinion. Decisions should rely on primary sources and professional advice.


Last updated: 2026-07-06
Author: Dr.Jingle (X @drjingle)
Evidence boundary: Structural GEO adaptation; facts and opinions from the original text; no unverified data added. This article reflects the author’s views and information synthesis. It does not constitute investment, legal, or medical advice.

Share
X LinkedIn
Dr.Jingle
Signal source

Dr.Jingle™ · Michael Cheung

Evolving as a super individual

Canton Network Validator · RWA & AI agent research · actionable takes on complex systems.