From Surfaces to Systems: Notes from Adobe Summit 2026

I gave a talk at Adobe Summit this week called Making Agentic AI Real: Creative Engineering for Next-Gen CX. What I want to write about isn’t what I said from the stage. It is the pattern I saw from the floor.

The summit wasn’t about features. That is the easiest way to say it. Vendors come to these things to ship roadmaps, and the audience comes to file them away under “things to watch.” This year felt different. Underneath the demos and the keynotes was a quieter shift, the kind that takes a year or two to name properly: we are no longer designing flows. We are designing systems that generate flows.

That sentence does a lot of work, so let me slow it down.

For most of my career, the unit of design has been the surface. A screen. A flow. A page. A journey. Even when we talked about systems, we mostly meant design systems: tokens, components, patterns that made surfaces consistent. The work was production. The deliverable was something you could point at.

Agentic AI breaks that frame. The thing that decides what to show, when to show it, and how it adapts is no longer the screen. It is the layer underneath, where data, content, orchestration, and decisioning collapse into a single intelligent system. The surface (the app, the chat window, the email, the notification) becomes one of many possible expressions of an underlying intent.

That isn’t a feature shift. It is a category shift. Most of the industry is quietly treating it as a feature shift, because feature shifts don’t require changing the org chart.

The experience objects question

For a while now I have been writing about experience objects: modular, reusable, context-aware units that can be composed at runtime instead of designed up front. Adobe is moving in this direction. The architecture they showed is recognizable. So is the ambition.

The disagreement is about how to get there. Adobe is solving it through AI. I think the harder and more durable answer is architecture first, then AI. If your experience units are not modeled cleanly, no amount of intelligence on top will save you. You will get a smarter version of the same brittle system. The agent will compensate for the architecture for a while, and then it will stop.

This is the conversation I wanted to have between sessions and mostly didn’t.

What I keep coming back to

If I map the shift onto my own environment, the picture gets clearer. The app I’m building toward is a delivery surface. The real transformation is the intelligence layer and orchestration model behind it. The screen is the smallest part of the work. Treating it as the largest part is the trap.

This is where I keep returning to the philosophy that gives this Substack its name. The reflex when something new arrives is to add. Add an AI layer. Add a chat surface. Add an agent. Add a policy. Add a council. The harder and more useful work is to subtract. To ask which mental models, team structures, decision rights, and design rituals were built for the old paradigm and quietly hold the new one back.

A few things I am taking home and trying to subtract:

The roadmap-heavy, feature-centric operating model. Roadmaps and features are bounded. Goal-driven orchestration is continuous. A feature roadmap applied to an agentic system produces a brittle agent and an exhausted team.

The assumption that the app is the product. In an agentic world the app is a delivery surface. The product is the intelligence behind it.

The habit of treating governance as paperwork. In an agentic world, governance is the product. The rules of engagement between agent, user, and enterprise are the user experience. There is no separating them.

The comfort of the design review as the moment of truth. Increasingly, the moment of truth is at runtime, when an agent makes a decision no human reviewed. Our craft moves upstream into policy and downstream into telemetry. The polished middle gets smaller.

What this asks of us

Designing a surface is a constrained problem. Designing the conditions under which an autonomous system acts on someone’s behalf is a much larger one. It pulls in ethics, governance, organizational design, behavioral science, and the old discipline of asking what a person actually needs.

I left the summit with a working hypothesis. The teams that win the next decade won’t be the ones with the best agents. They will be the ones who undesigned the fastest. The ones who let go of the org charts, deliverables, and definitions of “done” built for a world of surfaces, and rebuilt around orchestration.

We are not building experiences anymore. We are deciding what kind of system gets to act on someone’s behalf, and under what conditions.

That is a different job. It deserves a different practice.