AI agent dashboard design: the screens an agent platform needs

By Ingot UI3 min read

An AI agent dashboard is a control room, not a chat window. It needs five screens: an agent registry (what runs, who owns it, what it can touch), access policies, credentials with short lifetimes and rotation, a human-approval queue for risky actions, and an audit log of every decision. Put the numbers that signal risk in the top row: agents running now, over-privileged agents, requests blocked today and approvals waiting.

Summarize withChatGPTClaudePerplexity

On this page
  1. Why agent products need a different dashboard
  2. The five screens
  3. KPIs that signal risk
  4. Design details that matter for agents
  5. Mistakes to avoid

This guide is published by Ingot UI. Its Auton system was designed for exactly this kind of product, so the examples come from it: 279 components from 71 source screens, 264 of them with React code.

Why agent products need a different dashboard

A normal SaaS dashboard reports what people did. An agent platform has to show what software did on people’s behalf, with what permissions, and who allowed it. The questions change from “how many users?” to “which agent touched production data, under which policy, and did a human approve it?”.

The five screens

ScreenAnswersKey components
Agents (registry)What agents exist, who owns them, what they can reachAgents table, deploy drawer
Access policiesWhat each agent may do, in which environmentPolicies table, policy templates
CredentialsWhich secrets exist, how long they live, what is overdueCredentials table, credential health
ApprovalsWhat is waiting for a human, and how fast we answerApprovals page, approval requested
Audit logEverything that happened, filterable and exportableEvents table, export modal

KPIs that signal risk

Top-row numbers should make risk visible at a glance. Counts alone are not enough; pair each with a comparison or a threshold.

KPI, Running nowAutonCode Pro
KPI, Blocked todayAutonCode Pro
Auton KPI cards: running now, over-privileged, blocked today, average approval time.
  • Running now and registered: scale and activity.
  • Over-privileged agents: permissions wider than what they use. The most important risk number.
  • Requests blocked today: policies doing their job, or a misconfiguration.
  • Approvals waiting and average approval time: whether humans are a bottleneck.
  • Credentials expiring soon and rotation overdue: secrets hygiene.

Design details that matter for agents

  • Slow down risky actions on purpose. Deploys, credential issues and policy changes get their own drawer or confirm modal that names the agent and the scope.
  • Show scope inline. In tables, show environment and access scope as chips (Production, Read, +2) so reviewers see blast radius without opening the row.
  • Make time explicit. Credential lifetimes and “last active” matter more here than in most products; use relative times with exact values on hover.
  • Every event is linkable. Approvals and audit entries need stable URLs so they can be referenced in incident reviews.
  • Export is a feature. Audit exports (CSV, JSON, date range) are often required for compliance reviews.
Agents tableAutonCode Pro
Approval grantedAutonCode Pro
Events tableAutonCode Pro

Mistakes to avoid

  • Treating the chat transcript as the dashboard. Operators need state and permissions, not conversation history.
  • Hiding permissions behind a details click.
  • One “approve all” button for a queue of mixed-risk requests.
  • Audit logs without filters by agent, actor, action and time.

Frequently asked questions

What should an AI agent dashboard show?

The agents that exist and who owns them, what each may access, the credentials they use, actions waiting for human approval, and an audit log of everything they did.

What are the most important AI agent KPIs?

Agents running now, over-privileged agents, requests blocked by policy, approvals waiting with average approval time, and credentials expiring or overdue for rotation.

Is there a UI kit for AI agent dashboards?

Yes. The Auton system in Ingot UI covers agent registries, access policies, credentials, approvals and audit logs, with 264 components in React.

About Ingot UI

Ingot UI is a library of six dashboard design systems built from real product screens. Every generated React component is checked against its Figma original before it ships. How we write guides.

Components in this guide

Agents tableAutonCode Pro
Credentials tableAutonCode Pro
Events tableAutonCode Pro

Keep reading