A pattern library for responsible AI
Trust calibration patterns for AI product development
Teams building AI products sit between two worlds: compliance requirements, and design systems that say little or nothing about oversight. This library sits in that gap. Each pattern translates a governance obligation into a concrete design decision, the kind a product manager or UX designer can act on in a sprint, not a policy document to file away. If you want your AI design decisions to stand up to regulatory requirements, you're in the right place.
How to use this library
Product teams and designers can use the categories and oversight levels to find actionable patterns that promote transparency, build trust and support human oversight. This is a growing library.
- 1Navigate categoriesBrowse patterns grouped by theme (Trust and Transparency, Human Oversight, Explainability and Agent Behaviour) to address specific risk profiles.
- 2Assess risk & autonomyUse the Governance Calibration Tool to find the right level of human oversight (HITL, HOTL or HIC) based on decision consequences.
- 3Implementation checklistReview each pattern's “Questions to ask” at project kickoff to align requirements across cross-functional teams.
Why this library
Most AI guidance stops at principles. This library goes one step further: it names the failure mode, shows what it looks like in a real interface, and links it to the regulation that applies. Every pattern is grounded in the EU AI Act, ISO/IEC 42001 or my survey research, cited at the level of the design decision, not the page footer.
Use individual patterns to solve a specific design problem, or use the Governance Calibration Tool to find the right oversight level for your use case first, then browse the patterns that apply.
Built on 12 design principles for human oversight →
Add your voice. The practitioner survey behind this library is still open. Your answers shape the next patterns. Take the survey ↗
Resources used
Explore by category
How oversight is calibrated
Three ways people stay in charge of an AI system. Pick one to see where the person sits.
Human-in-the-Loop (HITL)
Serves as mandatory approval checkpoints for high-risk, irreversible actions. Here, the human acts as an explicit gatekeeper.
In practice: Hard Gate · Decouple Review Actions · Inline Evidence
Human-on-the-Loop (HOTL)
A monitoring-based approach where the AI operates autonomously while the user maintains supervisory oversight via dashboards and alerts.
In practice: Monitoring Dashboard · Fallback Mechanism · User Action Recovery
Human-in-Command (HIC)
A policy-level paradigm where users retain ultimate authority to define system boundaries, operational scopes, and safety constraints. HIC provides the underlying governance infrastructure, enabling users to set macro-level guardrails and override parameters before execution.
In practice: AI Disclosure · Audit Trail
Human oversight cannot be treated as a static control pattern; for instance, classifying an entire agentic workflow strictly as HITL when discrete sub-steps support HOTL monitoring. Instead, human agency serves as the governing constraint: AI autonomy remains subordinate to human authority, and users must retain the ability to challenge, overrule, or audit AI outputs.
Governance Calibration Tool · high-level proposal
Find the governance archetype for a workflow
To put governance into practice, use the Governance Calibration Tool concept below. Select a level for each factor, and it suggests the primary governance archetype (HITL, HOTL, HIC or Hybrid) for your workflow.
This is a high-level proposal. The result is a starting point for a team conversation about where human oversight is genuinely needed, not a risk assessment, legal review or compliance decision. Use the suggested patterns to turn it into design decisions, and revisit it as your workflow and its risks change.
Includes your selections, the suggested archetype, and the kickoff questions from each recommended pattern.
About the team
I am an experienced UX designer with expertise in agentic user research, product design and service design, and growing expertise in responsible AI. I work at the intersection of product and service design, turning responsible AI into actionable user experience patterns and improving how product and UX collaborate.
This library draws on my research into EU AI Act implementation, ISO/IEC 42001 and other AI policies applied to UX, as well as my survey of practitioners on how confidence signals, oversight and AI disclosure land in real products.