Agentic UX

A decision framework for how much authority an AI agent should have, and what the product must guarantee at each level.

Assistants answer questions. Agents take actions. A wrong answer is awkward; a wrong action can delete a file, spend money, or email a customer. Choose the right autonomy before you ship.

v2.0
Share

How much should the agent do?

L3

Confirm

Agent proposes. You say go.

YouAgent

Left = you stay in control. Right = the agent takes more of the work.

Autonomy is a product decision, not a model capability.

An agent may be able to take an action without being ready to receive authority over it. Separate what the model can do from what the product should allow, then match authority to consequence.

Design principle

Grant the lowest level of autonomy that delivers the outcome. Raise authority only when evidence shows the system can stay safe, legible, and recoverable.

When signals conflict, choose the lower level, or split the workflow so low-risk steps proceed while high-impact steps stay under review.

The five levels

Higher is not better. Each level changes who decides, acts, and monitors. Each section includes what to build and patterns that put it into the product.

Suggest

Question
Can the agent help without taking any action?
Principle
The agent never takes the action. It only surfaces options you can ignore.
Breaks as
A single pushy recommendation with no alternatives, or a suggestion that looks like a completed action.
Use when
Ideas, rankings, or next steps where a wrong suggestion is cheap to ignore.
Build
  • Show why this was suggested
  • Offer alternatives, not one answer
  • Let people dismiss it with no penalty
L1 decide, act, and monitor roles
RoleWho
DecidesYou
ActsYou
MonitorsYou

Pattern demo

Suggest / Confirm / Execute

Map suggest, confirm, and execute onto one product control so L1 stays advisory.

Open pattern

Draft

Question
Can the agent prepare work that only ships when a person finishes it?
Principle
The agent prepares an artifact, but nothing ships until you edit and intentionally complete it.
Breaks as
A draft that auto-sends, edits durable records, or hides that side effects are still blocked.
Use when
Emails, docs, code, or plans that must not go out until a person finishes them.
Build
  • Mark it clearly as a draft
  • Make every part editable
  • Block side effects until publish
L2 decide, act, and monitor roles
RoleWho
DecidesYou
ActsYou finish
MonitorsYou

Pattern demo

Chat Artifacts

Keep drafts editable and side-effect free until a person intentionally publishes.

Open pattern

Confirm

Question
Is the agent ready to act, but still paused until a person approves?
Principle
The agent is ready to act, but stays paused until you approve the exact scope.
Breaks as
Approve without a preview of impact, no deny or edit path, or silent re-plans after the user said yes.
Use when
Send, spend, delete, or change access: actions with real consequences that need a yes first.
Build
  • Preview exact impact before approve
  • Allow edit, deny, or cancel
  • Re-ask if the plan changes
L3 decide, act, and monitor roles
RoleWho
DecidesYou approve
ActsAgent after OK
MonitorsYou

Pattern demo

Human in the loop

The approval gate when the agent is ready to act but a person still decides.

Open pattern

Bounded

Question
Can the agent run alone inside hard limits and escalate on exceptions?
Principle
The agent runs on its own inside hard limits, and escalates when it hits an exception.
Breaks as
Soft limits the agent can ignore, no interrupt, or exceptions that fail silently without an owner.
Use when
Repeatable work where the agent can run alone inside budgets, allowlists, or timeboxes.
Build
  • Hard limits: budget, scope, stop rules
  • Escalate when uncertain
  • One-click interrupt and rollback
L4 decide, act, and monitor roles
RoleWho
DecidesYou set policy
ActsAgent in bounds
MonitorsYou on exceptions

Pattern demo

Per-Action Autonomy

Different actions in the same product can sit at different levels inside hard limits.

Open pattern

Monitor

Question
Can the agent operate continuously while people govern policy and anomalies?
Principle
The agent operates continuously. People govern policy and anomalies, not every step.
Breaks as
An always-on agent with no visible status, no anomaly owner, or no safe shutdown path.
Use when
Stable, well-owned operations where people oversee the system more than each action.
Build
  • Always-on status that it is running
  • Anomaly alerts with a clear owner
  • Safe shutdown and reauthorization
L5 decide, act, and monitor roles
RoleWho
DecidesYou set policy
ActsAgent
MonitorsYou + systems

Pattern demo

Autonomous Mode Display

When the agent runs without asking, that state has to be unmistakable.

Open pattern

Match the risk of one action to an autonomy level.

Score impact, reversibility, reliability, sensitivity, and detection. The tool recommends a level from Suggest to Monitor, plus who should decide, act, and monitor.

01 / Impact

If this goes wrong, how bad is it?

02 / Reversibility

Can you undo it?

03 / Reliability

How proven is the agent at this task?

04 / Sensitivity

How sensitive is the context?

05 / Detection

Can someone catch a bad outcome?

Cheatsheet

Eight checks before raising autonomy.

  • Is this the lowest level that still delivers the outcome?
  • Can people see who decides, who acts, and who monitors?
  • Are side effects blocked until the level allows them?
  • If the plan changes, does the product re-ask or re-bound?
  • Is there a one-click interrupt, deny, or shutdown path?
  • Are hard limits real (budget, scope, allowlist), not soft preferences?
  • When autonomy rises, is status unmistakable and owned?
  • Can someone recover work after a bad autonomous action?