The Rise of Design for Agents

Source: Hugh Dubberly; Dubberly Design Office, IxDC Beijing talk deck · 2025-10-31

This presentation focuses on how Large-Language Models (LLMs) can be used in software agents.

An axiom of design theory might be: Design practice changes as new technologies emerge.

After a relatively stable, normal period, technological change creates a “crisis”, which results in a new “paradigm”, which then becomes normalized. And the cycle repeats.

For millennia, design was an integral part of hand-craft production. Graphic design and product design subsumed most of hand-craft design.

The first design paradigm, circa ~1900: Mass manufacturing replaced most hand-craft production.

Graphic design and product design subsumed most of hand-craft design.

Interaction design subsumed large parts of graphic design.

Design for agents will subsume large parts of interaction design.

The third design paradigm, circa ~2025: LLMs are not only design tools but also a new material of design.

Services are processes, coordinating the actions of people and products. They are co-created by producers and consumers at the point of delivery.

Users will show up with their own agents as well.

Every service will offer software agents to assist consumers and providers — i.e., agent design is becoming core to service design.

Implied is that agents have rules for choosing actions; information, feedback, goals, teleology, cybernetics, learning, and even play are implicated.

What is an “agent”?

“An agent is anything that can be viewed as perceiving its environment through sensors and acting upon that environment through actuators.”

— Stuart Russell and Peter Norvig, Artificial Intelligence, A Modern Approach, 2010.

The capabilities of software agents will build in a stack of layers.

LLM-based chatbots are a simple form of agent. (e.g., in the US: ChatGPT, Claude, Cursor, and in China: ChatGLM, Ling-1T) They take an input and transform* it to return an output.

Agents often have a second input — a system prompt, describing the agent’s role, tone, and constraints.

Agents may draw on information sources outside the LLM through a process called Retrieval Augmented Generation (RAG).

When they grow up, agents will be everywhere — billions of them.

Every physical process and virtual process (that matters) will be overseen by a monitoring agent (a.k.a. Unix daemons), alerting supervisor agents when errors occur or are predicted.

The supervisor agents will ladder up in a management tree to an orchestrator agent at a human touch-point (though humans may drill all the way down if they like).

Xerox PARC researcher Rich Gold (1995) offered a dystopian counterview of “intelligent houses”: “How smart does your bed have to be, before you are afraid to go to sleep at night?”

All these agents will need to be designed — along with the touchpoints through which they interact, the processes that constitute our services, and the product-service ecologies that form our world.

This space of opportunities is part of the future of design.