Should agent customization be split into foundational and extended tiers?

Microsoft's Learn design guidelines for agents — a preview content set, page last updated 2026-07-14 — frame agent customization as two levels. "Foundational customizations, like an agent's icon and name, establish its identity and set user expectations," while "Extended customizations, like dialogs and stream responses, define the quality of the interaction itself." The guidelines ground this split in three pillars that "should inform every design decision": Responsible AI ("transparent, fair, and worthy of user trust"), Accessibility and inclusion ("works for users of all abilities and backgrounds"), and Human-centered design ("the user's needs, context, and goals at the center of your process"). The same content set also treats an agent's "visual anatomy" — its core UI components — and its "entry points," the ways users discover and access it, as foundational design knowledge that precedes any customization decision.

The guidelines' reasoning is that identity and interaction quality are different design problems sitting at different points in an agent's lifecycle. Foundational choices (icon, name) are made once and shape a user's first expectations of what the agent is and can do; extended choices (how it holds a dialog, how it streams a response) are made continuously and shape how the interaction actually feels once underway. Microsoft presents the tiering as a sequencing device for designers: establish identity first, then layer behavior on top of it.

The foundational/extended split is a different organizing axis than other agent-design frameworks already in the library. How should AI systems be designed across different interaction phases? organizes guidance by when in a session a problem occurs; Microsoft's guidelines instead organize by what kind of customization decision a designer is making, independent of session phase. It also differs from What design areas matter most for computer use agent UX?, a bottom-up taxonomy built from studying existing agents and interviewing practitioners, where Microsoft's three pillars and two-tier model read as top-down design doctrine issued by the platform vendor rather than derived from a study. It overlaps loosely with What UX principles do workplace users want in AI agents? in naming trust and user-centeredness as core concerns, but Microsoft's pillars are stated as values to design toward, not findings about what users in a study weighted most heavily.

The excerpt is an overview page for a larger guide, not a study — it states the framework's categories and a few illustrative examples (icon, name, dialogs, streaming responses) but does not show how the foundational/extended split was arrived at, nor does it report any validation with designers or users. The guidance should be read as one vendor's prescriptive design doctrine for its own agent-building context, not as evidence that this particular split improves design outcomes.