What design skills do teams need for the AI era?

In Salesforce Design's 2025 retrospective, Hsiao-Ching Chou reports that Chief Experience Officer Kat Holmes "implored CEOs to invest in their teams and focus on four critical design skills to thrive with AI: experience architecture, designing agent collaboration, conversation design, and building context maps and ontologies." The four are presented as a call to action for leadership, not a product requirement list — a statement of what design teams need to be good at, rather than what an AI interface needs to contain.

Each of the four skills maps onto a gap the same retrospective describes elsewhere. Experience architecture is the work Design Director JoEllen Kames describes as building the "'invisible' scaffolding that enables AI systems to respond in real time" so they "surface the right things at the right time and in the right places." Designing agent collaboration is the gap Principal Product Designer Hervé Mischler names directly — "traditional UX design doesn't have frameworks for coding assistants and multi-agent orchestration" — and the one Principal Design Strategists Andrea Small and Kate Piper, with Experience Strategy Architect Bethany Pickard, frame as a race against a status quo where "most tools tend to support independent work," pushing toward "'multiplayer' experiences that enable teams to collaborate more easily with one another and AI agents." Conversation design covers the retrospective's own thread on "what is conversation design," "how to repair broken conversations between humans and AI," and the effect of conversation design on the quality of a call with an AI agent. And context maps and ontologies is grounded by Principal Ontologist Madonnalisa Chan's claim that ontology — "a system for organizing the data about the data" — is why AI "understand[s] the language and context of your business" at all: "without ontologies, your experience using any number of digital platforms, such as Netflix or Spotify, would be chaotic."

Against the nearest guideline-style notes, Holmes's four skills are organized by design discipline rather than by interaction phase or end-user principle. How should AI systems be designed across different interaction phases? groups its recommendations by when during the user's interaction they applied — initially, during, when wrong, over time — describing what an interface should do. Holmes's four instead describe what a design team should learn to do, which is a competency roadmap rather than an interface specification; the two could coexist, with Amershi's phases describing interaction-level requirements that experience architecture and conversation design, as skills, are meant to satisfy. This is also the skills framework behind Has AI work moved designers upstream from screens?, the broader screens to scaffolding shift this same retrospective describes — Holmes's list is one way the retrospective operationalizes that shift into named, learnable skills.

The excerpt gives one sentence naming the four skills, attributed to a single executive's internal call to Salesforce's own leadership; it is not a validated competency model, and the retrospective offers no evidence — no study, no practitioner survey, no before-and-after comparison — that these four skills predict better AI products or that design organizations which build them outperform those that do not. The skills are also named inside one company's internal vocabulary (tied to Agentforce and Salesforce's own design practice) rather than demonstrated as a general taxonomy. Read at the strength the excerpt supports, this is a checklist of vocabulary one design organization is using to describe what AI-era work requires, not evidence that these four skills are necessary or sufficient.