Mandate Design Patterns

The essay identifies five structural patterns that every mandate system must implement. Intent capture addresses how humans express what they want an agent to do, offering three modes: conversational capture for complex or novel mandates, structured capture using form-based interfaces for routine tasks, and template-based capture for high-frequency delegations. The most effective systems combine all three, allowing a user to start from a template, adjust constraints through structured inputs, and handle exceptions conversationally. Boundary specification defines the operational envelope through four constraint categories: financial boundaries (budget ceilings, per-item limits), temporal boundaries (expiry times, blackout periods), quality boundaries (brand preferences, ethical constraints), and scope boundaries (permitted merchants and categories).

Authority gradients calibrate the spectrum of autonomy a mandate grants, operating across four levels: observation (monitor and report), recommendation (research and suggest), conditional autonomy (act within parameters, escalate outside them), and full autonomy (act independently within boundaries). The essay argues that authority should vary within a single mandate depending on the decision type - a grocery agent might auto-approve familiar items but escalate unfamiliar substitutions. Lifecycle management governs the mandate from creation through activation, monitoring, amendment, and completion, with each phase requiring distinct interface treatment. Every mandate must have an explicit expiry to prevent what the essay terms "zombie mandates" - standing authorisations that persist beyond their intended purpose.

The fifth pattern, failure and escalation, governs what happens when an agent encounters conditions outside the mandate's parameters. The essay distinguishes constraint violations (budget insufficient), ambiguity (mandate does not address the situation), and mandate conflicts (contradictory active mandates). Five anti-patterns recur across implementations: the blank cheque mandate (over-broad authority), the interrogation mandate (excessive specification demands), zombie mandates (forgotten active authorisations), opaque mandates (technically precise but humanly meaningless constraints), and static mandates (cannot be amended once activated). In agentic commerce, these patterns map directly to the Agent Payments Protocol (AP2) three-tier credential system of Intent Mandate, Cart Mandate, and Payment Mandate.

For product teams building agentic systems, the mandate interface is the primary site of trust architecture work. A poorly designed mandate produces vague, over-broad authorisations that erode trust when agents act outside user expectations. A well-designed mandate produces precise, bounded, reviewable contracts that give humans genuine control. The essay's implementation principles are direct: start with templates rather than blank canvases, make boundaries visible rather than hidden, use progressive disclosure to match interface complexity to mandate complexity, and always render a plain-language summary before activation.

Designers must treat amendment and audit as first-class concerns, not afterthoughts. Users must be able to modify active mandates and see the implications of changes before confirming them. Every agent action must be traceable to the mandate that authorised it, enabling the post-execution review through which users calibrate their mandates over time. In regulated domains such as financial services, this audit trail is a compliance requirement. The mandate is the foundational interface of the agentic age - the mechanism through which delegation design becomes operational and human oversight becomes structural rather than aspirational.

Organisations that treat mandate design as a core product capability—rather than a compliance checkbox—will build the trust architecture that earns sustained human delegation in agentic commerce and agentic shopping.