Has AI work moved designers upstream from screens?

Salesforce Design's year-end retrospective opens with the claim that "design sets the foundation for trustworthy AI" because "designers help establish the semantic foundation that teaches AI what things mean and how humans work." Hsiao-Ching Chou, writing for Salesforce Design, frames the throughline of the company's 2025 design work as a shift in what designers actually do: "to borrow the words of a colleague who's a principal AI architect, design used to be about creating screens. Now, it's about deciding what happens before the screen appears." Kat Holmes, Salesforce's Chief Experience Officer, names this shift "agent experience design, where AI agents are treated as a new kind of user."

The reasoning the piece gives is that AI systems are "dynamic" and have "no need to have a single user journey or click path," so the old unit of design work — a screen, a flow — stops being where the leverage is. Design Director JoEllen Kames describes product designers having to architect "the 'invisible' scaffolding that enables AI systems to respond in real time" and to "create experiences that surface the right things at the right time and in the right places." Principal Product Designer Hervé Mischler makes the same point from the gap it leaves: "traditional UX design doesn't have frameworks for coding assistants and multi-agent orchestration," which is why designers must build "experiences that harness coordinated AI agents across platforms while maintaining human agency and understanding." Former Salesforce designer Min Chang generalizes it as systems thinking: "instead of designing for isolated screens, I'm now designing how a system moves between states."

This gives a practitioner-side name to work that How should AI interfaces handle the shift from doing to supervising? formalizes into an explicit architecture: Nielsen's Orchestration Surface, the layer where an agent reveal[s] its proposed plan and seeks consent before acting, is a concrete design answer to the same "before the screen appears" territory Chou's colleagues describe informally, and his Direct-Manipulation Surface is the GUI that Salesforce's designers say is no longer the primary unit of design. Against that backdrop, How should AI systems be designed across different interaction phases? is the guideline set this piece implicitly argues has been outgrown: Amershi's guidelines still assume a graphical interface to correct and calibrate, where Chou's retrospective describes designers working one layer upstream of that interface, on orchestration and semantics rather than on-screen correction.

What the excerpt does not establish is whether this shift actually improves outcomes for users or businesses. It is a company blog's retrospective, assembling quotes from named Salesforce employees and executives about their own year of work — not a study, and not independently validated: there is no usability data, adoption figure, or comparison showing that "agent experience design" or "invisible scaffolding" work produces better agent experiences than screen-first design would. The piece is also specific to one vendor's product context (Salesforce's Agentforce platform and Lightning-based SLDS 2 design system), so the generalization to design practice outside Salesforce is an extrapolation the source itself does not make. The implication, held at that strength, is that this vocabulary shift — from screens to scaffolding — describes how one company's design organization narrates its own 2025 work, not yet a demonstrated change in what makes AI products usable.