← Glossary
Agentic Experience (AX)
Agentic Experience (AX) is the design of human–AI collaboration when software can plan, decide, and act—often across tools, data, and policies. It’s less “chat UI” and more workflow orchestration: who does what, with what authority, and how outcomes stay contestable.
Definition
- AX is the end‑to‑end experience of a user + an agent (or agent network) completing a job: intent → plan → actions → confirmation → audit trail.
- It includes the interaction layer (what the user sees), but also the control layer: autonomy bounds, handoffs, reversibility, and evidence surfaces (“why this”).
- AX is successful when a system can act with speed and remain understandable under stress.
Why it matters
- Agents collapse UI into outcomes. Users care less about where a button lives and more about whether the system acted correctly, safely, and transparently.
- In regulated or high‑stakes contexts, “helpful” is not enough. You need traceability, recoverability, and operator‑grade oversight.
- AX is where product, design, security, legal, and operations converge. Without a shared model, teams ship demos—not capability.
Common failure modes
- Autonomy without gradients: either everything is manual (no value) or everything is automatic (risk).
- No state model: the agent “did something”, but nobody can explain where the process is, or how to recover.
- Silent failure: the system fails quietly, then users discover damage downstream.
- Explainability theatre: generic “because AI said so” messages, no evidence trail, no ability to challenge.
- No escalation: when uncertainty is high, the agent still proceeds—because there is nowhere to hand off.
How I design it
- Start with the job and the states, not the UI. Define: requested → in progress → awaiting input → actioned → pending confirmation → completed, plus failure/rollback states.
- Design a progressive autonomy ladder: suggest → draft → execute with confirmation → execute with post‑hoc notification → fully autonomous (rare).
- Make handoff and override a first‑class flow. Define who can intervene, when, and what happens next.
- Surface evidence: inputs, policy constraints, and reasoning signals that a human can evaluate quickly.
- Instrument for operational truth: logs that operators can use, not just machines.
Related work
Proof map claims
Case studies
See also
Contact
Let’s discuss a leadership role, advisory work, or a complex product challenge.