Adobe Principal Scientist Carlos Sanchez presented a concept at the AI Engineer World’s Fair (AIEWF) in San Francisco in which a website is assembled in real time around the needs of each visitor. Rather than selecting from a fixed set of prebuilt personalization options, the approach interprets visitor intent, retrieves relevant company content, and composes a tailored page on the fly.
Sanchez called the approach an "agentic site" and summarized Adobe’s framing as an "audience of one": the site personalizes itself in real time for the specific user and their current actions.
How the system works
In demonstrations, Sanchez showed a site that used browsing behavior and search queries as input signals. The system grouped those signals into intent categories — for example browsing, researching, or preparing to purchase — and then used a large language model (LLM) to assemble a page matched to that intent.
One demo presented a coffee‑machine site adapted for a visitor interested in camping: the copy, product selection, and supporting content were reorganized to emphasize making coffee outdoors. Another interface allowed an open query such as “Europe AI conferences” and returned a page composed specifically for that request.
Content grounding and retrieval
Adobe’s system retrieves material from the company’s existing content corpus (grounding corpus) rather than asking an LLM to invent an entire experience from scratch. That retrieval approach anchors the generated page to actual existing content.
Performance, cost and deployment status
Sanchez said Adobe evaluates models for both accuracy and latency: the company prefers that site generation take no more than one to two seconds. He estimated current inference cost at around one to two cents per page and noted that costs are likely to decline over time. Adobe has not broadly deployed agentic sites to production customer sites yet; the company is presenting the idea to customers and seeking organizations willing to experiment.
Use cases and uncertainty
E‑commerce is an obvious early use case because personalization can be tied directly to conversion, but Sanchez argued the approach could work wherever there is a large matrix of user types or personas. He also cautioned that it is not yet clear which agentic experiences customers will adopt. “With AI, it’s very easy to build things, but it’s hard to know what to build,” he said. “We build things and then we find the customers.”
Human visitors and personal agents
Sanchez expects websites to evolve to serve both human visitors and autonomous personal agents. As agents become capable of carrying richer expressions of user preferences and performing tasks autonomously, they may request or carry out purchases on behalf of users. That implies sites must support varying levels of delegation and involvement — for example, an agent might reorder basic supplies autonomously, while a human may still want a visual interface to choose a jacket.
Integration technologies
Technologies such as WebMCP could allow a site to expose structured tools directly to an agent, MCP Apps and other generative interfaces could bring interactive product experiences into a chat environment, and an A2A (agent‑to‑agent) backend might let agents interact without traversing the conventional visual site.
What this means for the web
Sanchez suggested the traditional website is unlikely to disappear entirely, but its role will change: instead of being a fixed collection of pages, a website may become a governed content and interaction system that assembles an appropriate interface on demand. Adobe is actively exploring that future while customers and marketers work out how to combine chat, UI and agentic approaches in practice.
Summary
Adobe’s "audience of one" demonstrations show that generating personalized pages in real time is technically feasible today, with latency targets of one to two seconds and estimated inference costs of one to two cents per page. The concept remains experimental: Adobe is seeking customers to pilot agentic sites, and the industry is still determining which combinations of human‑ and agent‑facing experiences will gain traction.
