The Website of the Future May Assemble Itself for Every Visitor

TL;DR

• Adobe is pioneering "agentic sites" that use AI to dynamically assemble web experiences tailored to individual visitors in real-time, moving beyond static templates and basic personalization.
• This isn't just personalization 2.0 — it's a fundamental architectural shift where websites become composable systems that reason about user intent and reconstruct themselves accordingly.
• Product builders must rethink their tech stacks now: traditional CMS platforms, A/B testing frameworks, and analytics tools weren't designed for experiences that literally change shape for every session.
• The transition will be messy but inevitable — expect a 3-5 year window where hybrid approaches dominate, with progressive enhancement of existing sites rather than wholesale rebuilds.

---

The static website is dying. Not the way Flash died — with a dramatic sunset and universal mourning — but slowly, inevitably, like a star collapsing into a white dwarf. We've been watching this happen for years through progressive personalization: recommendation engines, dynamic pricing, geolocation-based content. But Adobe's recent exploration of "agentic sites" suggests we're approaching an inflection point that will make previous personalization efforts look quaint.

The concept is deceptively simple: websites that assemble themselves uniquely for every visitor, using AI agents to reason about user intent, context, and goals in real-time. But the implications for how we build, measure, and think about digital products are anything but simple.

Beyond Personalization: Understanding Agentic Architecture

Let's be clear about what we're discussing here. Agentic sites aren't just smarter versions of the recommendation algorithms that suggest products on Amazon or the content feeds that dominate social media. Those systems optimize within predefined structures — they might change what you see, but not how you see it.

Agentic sites go further. They reconstruct the interface itself based on inferred user intent. Imagine landing on an e-commerce site where the navigation, layout, content hierarchy, and even the fundamental user journey adapt to whether you're a first-time browser, a comparison shopper, or a returning customer with a specific purchase intent. The site doesn't just show you different products; it becomes a different site.

This is possible because of three converging technologies:

Component-based architecture has matured to the point where websites are genuinely composable systems. Modern frameworks treat UI elements as modular building blocks that can be assembled programmatically. This isn't new, but it's now ubiquitous.

Large language models can reason about user intent from minimal signals — search queries, click patterns, even the time spent hovering over elements. They can make sophisticated inferences about what someone is trying to accomplish.

Edge computing enables this reasoning to happen in milliseconds, not seconds. The latency problem that plagued earlier personalization attempts is largely solved.

When you combine these capabilities, you get something qualitatively different from what came before. You get sites that think.

My Take: This Changes Product Management Fundamentally

I think we're underestimating how disruptive this will be to product management practice. Here's why: the entire discipline evolved around the assumption that you ship a thing — a defined experience that all users encounter. Yes, we've had user segments and feature flags and personalization rules, but these were variations on a canonical version.

Agentic sites eliminate the canonical version. There is no single "product" to manage in the traditional sense. Instead, you're managing a system that generates products dynamically. This shifts product management from design and specification toward something closer to constraint definition and quality assurance.

You're no longer asking "What should this page look like?" You're asking "What are the boundaries within which the AI can operate? What are the invariants that must hold across all generated experiences? How do we measure success when every user sees something different?"

This is a profound change, and I don't think most product organizations are prepared for it. The tools, processes, and mental models we've built over the past two decades assume a level of control and predictability that agentic systems simply don't offer. We're going to need new frameworks, and we're going to need them soon.

The Technical Architecture: What Builders Need to Know

If you're building products today and want to be ready for this shift, here's what your technical foundation needs to support:

Composability at Every Layer

Your content, components, and even business logic need to be granular and recombinable. This means:

• Content as structured data, not page-level blobs. Every piece of content should have rich metadata about its purpose, audience, and relationships to other content.
• Truly headless architecture where your backend exposes capabilities, not pages. APIs should return components and data, not rendered HTML.
• Design systems that think in primitives, not templates. Your component library should be flexible enough to support arrangements you haven't explicitly designed.

This is a heavier lift than it sounds. Most organizations have accumulated years of technical debt in the form of page-centric architectures, tightly coupled content and presentation, and design systems that assume specific layouts.

Real-Time Inference Infrastructure

You need the ability to run sophisticated AI models at the edge with minimal latency. This means:

• Edge-deployed models for instant inference, not round-trips to centralized AI services.
• Efficient prompt engineering that can convey user context and retrieve relevant information in a single digit milliseconds.
• Fallback mechanisms for when inference fails or takes too long — the system needs to degrade gracefully to a reasonable default experience.

The good news is that edge AI infrastructure is rapidly maturing. Cloudflare Workers AI, AWS Lambda@Edge, and similar services make this increasingly accessible. But you still need to architect for it deliberately.

Observability for Non-Deterministic Systems

This is perhaps the hardest problem. When every user sees a different experience, how do you:

• Monitor for problems? Traditional error tracking assumes you can reproduce issues. With agentic sites, you might never see the exact combination of factors that led to a poor experience.
• Measure performance? Conversion rate optimization assumes you can compare variants. When there are infinite variants, what does "better" even mean?
• Debug issues? Users will report problems with experiences you didn't explicitly create. How do you trace back through the AI's decision-making process?

You'll need observability tools that can:

• Capture the full context of each generated experience
• Cluster similar experiences for pattern detection
• Track outcomes at the individual session level
• Provide replay capabilities that reconstruct what the AI actually generated

Most existing analytics platforms aren't built for this. Expect a new generation of tools to emerge specifically for agentic systems.

The Transition Path: Hybrid Approaches

Despite the revolutionary potential, I don't expect wholesale adoption of fully agentic sites in the near term. The transition will be gradual, with most organizations taking a hybrid approach:

Phase 1: Agentic components within static sites. Start with high-value, contained experiences — product recommendations, search results, onboarding flows — where the AI assembles components dynamically while the surrounding site remains static.

Phase 2: Agentic sections. Expand to entire sections of the site that operate agentically — perhaps the entire shopping experience becomes dynamic while marketing pages remain static.

Phase 3: Fully agentic sites with static fallbacks. The entire site is generated dynamically, but with well-defined fallback experiences for edge cases, accessibility requirements, or when the AI system is unavailable.

This phased approach lets you build organizational capability gradually. You'll learn how to design for non-determinism, how to QA experiences you can't fully predict, and how to measure success in a world of infinite variants.

The Design Challenge: Crafting Constraints, Not Experiences

For designers, this shift is particularly challenging. The craft of design has always been about making specific choices — this color, that typography, this layout. Agentic sites require designers to work at a higher level of abstraction.

You're no longer designing screens; you're designing systems that generate screens. This means:

Defining design languages, not designs. Your output is a set of principles, components, and rules that the AI uses to generate specific experiences. Think of it as writing a constitution rather than passing laws.

Designing for ranges, not points. Instead of specifying that a hero image should be 1200px wide, you might specify that it should occupy 40-60% of the viewport width depending on content importance and user context.

Embracing emergent aesthetics. Some of the experiences the AI generates will surprise you. They might not match your initial vision, but they might work better for specific users. You need to be comfortable with this emergence.

This requires a different skill set than traditional digital design. It's closer to systems thinking, to architecture, to designing processes rather than artifacts. Not every designer will make this transition easily.

Business Model Implications

Agentic sites also raise interesting questions about business models and competitive dynamics:

Agencies and consultancies that sell website builds as projects will need to evolve toward ongoing optimization services. You can't deliver an agentic site and walk away; it requires continuous refinement of the AI's decision-making.

CMS vendors are in a particularly interesting position. The ones that successfully pivot to supporting agentic architectures will thrive. Those that remain wedded to page-based content models will struggle. Adobe's early exploration of this space suggests they understand the stakes.

Competitive differentiation becomes more complex. When sites can adapt to individual users, the advantage goes to companies with better AI, better data about user intent, and better component libraries — not necessarily better static designs.

The Risks We Need to Talk About

I'd be remiss not to address the potential downsides of this technology:

Filter bubbles on steroids. If every site shows you a version optimized for your existing preferences and behaviors, we risk creating even more fragmented realities. The site that challenges your assumptions or exposes you to new ideas might be optimized away.

Manipulation at scale. An AI that can reconstruct a site to maximize engagement or conversion for you specifically is also an AI that can manipulate you specifically. The ethical frameworks for responsible use of this technology are still nascent.

Accessibility challenges. Screen readers and other assistive technologies assume some level of predictability in site structure. When that structure is generated dynamically, ensuring accessibility becomes significantly harder.

The death of serendipity. Some of the best web experiences come from unexpected discoveries — stumbling across content or features you weren't looking for. Over-optimization for intent might eliminate these happy accidents.

These aren't reasons to avoid the technology, but they are reasons to develop it thoughtfully, with guardrails and ethical considerations baked in from the start.

What Product Builders Should Do Now

If you're building digital products today, here's my advice for preparing for this shift:

Audit your architecture for composability. Can your content and components be easily recombined? If not, start refactoring toward more modular systems. This pays dividends even if agentic sites don't materialize exactly as predicted.

Experiment with constrained agentic experiences. Pick a small, contained part of your product and try building it as an agentic system. Learn the challenges in a low-risk environment.

Invest in your design system. A robust, flexible design system is the foundation of any agentic site. If yours is brittle or incomplete, prioritize fixing that.

Rethink your metrics. Start developing measurement frameworks that can handle non-deterministic experiences. What are the invariants you care about regardless of the specific experience generated?

Build AI literacy in your team. Product managers, designers, and engineers all need to understand how AI systems make decisions, what their limitations are, and how to design for them.

The Timeline: Sooner Than You Think

How fast will this happen? My guess is that we'll see production agentic sites from major brands within 18-24 months, with broader adoption following over the subsequent 3-5 years. This might sound aggressive, but consider:

• The underlying technologies are already mature
• The competitive pressure is intense — the first movers who get this right will have significant advantages
• The infrastructure providers (Adobe, Contentful, Vercel, etc.) are already building the tools

This isn't science fiction. It's engineering work happening right now.

The website of the future won't be something you visit; it will be something that assembles itself for you, in that moment, based on who you are and what you're trying to accomplish. For product builders, the question isn't whether this will happen, but whether you'll be ready when it does.

The static web had a good run. It's time to start building for what comes next.