Is design for agents becoming the next major design paradigm?

Hugh Dubberly, of Dubberly Design Office, argues in an October 2025 IxDC Beijing talk deck that design is entering a third paradigm, and that this paradigm is "design for agents." He frames the move as following an axiom of design theory: "Design practice changes as new technologies emerge," where "a relatively stable, normal period" gives way to a "crisis" that produces a "new paradigm," which then normalizes before the cycle repeats. The lineage he traces runs from hand-craft production, through mass manufacturing (circa 1900) subsuming most hand-craft design into graphic and product design, through interaction design subsuming "large parts of graphic design," to a third paradigm, circa 2025, in which "design for agents will subsume large parts of interaction design." He attributes the shift to large language models (LLMs) acting not only as design tools but also a new material of design.

The reasoning runs through services rather than products: Dubberly describes services as "processes, coordinating the actions of people and products," co-created by producers and consumers "at the point of delivery." Because "users will show up with their own agents as well," he argues "every service will offer software agents to assist consumers and providers," concluding that "agent design is becoming core to service design." He grounds the term "agent" in Russell and Norvig's definition — "anything that can be viewed as perceiving its environment through sensors and acting upon that environment through actuators" — and lists what designing agents implicates: rules for choosing actions, information, feedback, goals, teleology, cybernetics, learning, and play.

The claim sharpens Does agentic AI succeed through process redesign or workflow layering? by pushing the redesign argument up a level: where Deloitte argues enterprises must redesign processes rather than layer agents onto existing workflows, Dubberly argues the discipline doing that redesigning — interaction design — is itself being subsumed by a successor paradigm, design for agents. It also reframes Can AI agents truly operate above the loop at enterprise scale?: McKinsey describes humans repositioned above or within AI-first workflows as an organizational design; Dubberly treats that repositioning as a design-practice question, with service design absorbing agent design the way interaction design once absorbed graphic design. The companion note, How should agent capabilities be organized into functional layers?, gives the layered agent architecture this paradigm shift is meant to cover.

The deck is a conference talk's set of assertions and an explicit axiom, not a study of design practice; it offers no survey of studios or firms already practicing "design for agents," no adoption figures, and no criteria for when a paradigm is considered to have taken hold. The argument rests on an analogy to two prior shifts (mass manufacturing, interaction design), not on evidence that the current shift is underway at comparable scale. The claim should be read as Dubberly's extrapolation from design history, holding only as strongly as that two-case analogy supports.