Agentic AI Experience Design
Designing AI that can move work forward without removing human control.
I design agentic AI experiences around real workflows, decisions, permissions, and exceptions—helping intelligent systems understand context, recommend next steps, take appropriate action, and keep people informed and in control.
The shift from answering questions to helping complete work.
The most important design question is not what AI can do. It is what the AI should be allowed to do.
Traditional AI experiences often wait for a user to ask a question. Agentic experiences can interpret context, reason across information, use tools, recommend actions, and move a workflow forward.
That changes the role of UX. Designers must define how autonomy, permissions, confidence, approvals, errors, explanations, and escalation work together so users understand what is happening without having to manage every individual step.
Designing the relationship between people, agents, data, and action.
Agentic UX extends beyond conversational interfaces. It requires understanding the full operational system: what the agent knows, what it can access, what it can change, when it should act, and when people need to step in.
Give the agent enough context to understand the situation.
Effective agentic experiences depend on relevant user, workflow, system, business, and historical context rather than treating every interaction as an isolated request.
Help the system determine what should happen next.
I design around goals, dependencies, business rules, exceptions, available data, and decision points so an agent can move beyond simple responses toward useful next steps.
Define what the agent can do—not just what it can say.
Agentic systems may search records, generate documents, update systems, trigger workflows, coordinate tools, or complete tasks. Each action needs clear permissions and consequences.
Put human judgment where it matters most.
I define where agents can proceed independently, where users should review recommendations, and where explicit approval is required before an action changes data or affects an outcome.
Design for what happens when the agent is uncertain or wrong.
Strong agentic UX includes interruption, correction, undo, escalation, clarification, and recovery patterns—not just the ideal path where automation works perfectly.
Autonomy should increase only when confidence and risk allow it.
Not every agent action requires the same level of oversight. Low-risk, reversible actions may happen automatically while consequential or irreversible actions require stronger human involvement.
I design levels of autonomy around risk, confidence, reversibility, permissions, and the consequences of getting something wrong.
The agent proposes the next step.
The system analyzes context and presents a recommendation while the user remains responsible for taking action.
The agent prepares the action for approval.
The system determines and prepares an appropriate action but requires explicit confirmation before completing it.
The agent acts within defined boundaries.
Approved low-risk actions can happen autonomously with clear visibility, auditability, interruption, and recovery options.
Designing the full loop from intent to outcome.
Interpret the user's goal and surrounding context.
The experience should help the agent understand intent, current state, relevant history, available information, constraints, and what successful completion looks like.
Determine an appropriate path toward the goal.
The agent evaluates information, rules, dependencies, available tools, risk, and uncertainty to determine what action or sequence of actions may move the workflow forward.
Execute actions using the right systems and tools.
Agents may retrieve information, update records, generate artifacts, initiate workflows, coordinate systems, or complete authorized tasks on the user's behalf.
Keep people aware without exposing unnecessary complexity.
Users should understand what the agent did, what changed, what needs attention, and what happens next without having to interpret every internal step.
Use outcomes and feedback to improve future interactions.
Corrections, approvals, rejected recommendations, successful actions, and user feedback can inform how the experience handles similar situations in the future.
Intelligent action requires intentional boundaries.
Agents should only access systems, information, and actions that are appropriate for the user, task, and business context.
People need visibility into important agent actions, outcomes, assumptions, and changes without being overwhelmed by system complexity.
Uncertainty should influence whether an agent proceeds, recommends an action, requests clarification, or escalates to a person.
Wherever possible, autonomous actions should support undo, correction, cancellation, or recovery when the result is not what the user expected.
Agents need clear pathways for situations that exceed their permissions, confidence, available information, or decision boundaries.
Agent activity should remain understandable and traceable so teams can review what happened, why it happened, and what needs attention.
The interface is only one layer of the experience.
Designing agentic AI means understanding the entire system around the interaction: goals, context, tools, permissions, decisions, actions, feedback, and the points where human judgment becomes essential.
Define what the user is trying to accomplish rather than reducing the experience to individual commands.
Determine what information, history, system state, and business context the agent needs.
Identify what systems, data sources, services, or actions are available to complete the work.
Define permissions, autonomy levels, risk thresholds, approval requirements, and prohibited actions.
Communicate progress, decisions, changes, outcomes, uncertainty, and anything requiring user attention.
Provide ways to correct, undo, interrupt, clarify, escalate, and recover when the agent cannot safely continue.
The best agentic experiences do more than automate tasks. They reduce uncertainty.
I combine user research, systems thinking, workflow analysis, interaction design, AI experience design, accessibility, and human-in-the-loop principles to design intelligent products that can act with purpose while keeping people informed and in control.
