AI Pilot Modes: Ask, Plan, and Agent
Who this is for
Anyone using AI Pilot for the first time, or anyone wondering why the AI gives advice instead of acting.
What you will complete
Understand the three interaction modes of AI Pilot, when to use each, and what to expect from each.
Before you begin
- At least one cloud account must be connected.
- AI Pilot must be enabled under Settings → AI Agent → Agent Controls.
- Write actions require Owner or Admin role.
Overview
AI Pilot operates at three distinct levels of engagement. Understanding which applies to your request helps you work faster and avoids confusion when the AI gives advice instead of taking action.
Mode 1: Ask (Conversational)
The AI reads your server metrics, site statuses, alert history, and recent operations — but changes nothing. It responds with analysis or recommendations.
Use Ask mode for:
- Understanding metrics ("What does load15m mean and is mine too high?")
- Getting advice before a decision ("Which region should I use for my next server?")
- Diagnosing a problem you are investigating yourself ("My site returns 502 errors — what should I check?")
What you will see: A text response. No approval card. Nothing changes in your infrastructure.
Mode 2: Plan (Proposal review)
The AI describes a multi-step operation, shows you exactly what it intends to do, and waits for your approval before touching anything. Each write step gets its own approval card.
Use Plan mode for:
- Complex workflows (provision a server, deploy a site, configure SSL)
- Understanding the full impact before an operation starts
- Giving the AI a goal and letting it work out the steps
What you will see: A numbered plan, then individual approval cards for each write action. The AI proceeds only after you confirm.
Mode 3: Agent (Autonomous execution)
The AI executes approved actions sequentially and reports results at each step. You approve the plan once and the AI completes the full sequence.
Use Agent mode for:
- Routine trusted operations (restart a service, create a backup, renew SSL)
- Batch operations across multiple servers
What you will see: Actions run one after another. The Activity Center shows each operation progressing. The AI reports success or errors at each step.
Step-by-step: using AI Pilot
- Navigate to AI Pilot in the left sidebar.
- Type your question or instruction in the chat input at the bottom of the screen.
- Press Enter or click the send button.
- Watch the response type:
- Questions and analysis → text response only (Ask mode)
- Action requests → plan summary appears, then approval cards for write operations
- Review any approval card carefully. Read the action description and the impact level (Safe, Caution, or Destructive).
- Click Allow to approve a proposed action, or Deny to cancel it.
- For Destructive actions, type the server or site name shown in the confirmation prompt before Allow becomes active.
- The AI executes the approved action and reports the result in chat.
What success looks like
- Questions produce a clear analysis response within a few seconds.
- Action requests produce an approval card before any change is made.
- Nothing in your infrastructure changes without an approval card having appeared first.
Common errors and fixes
"AI Pilot shows as disabled" Cause: The master AI switch is off. Fix: Go to Settings → AI Agent → Agent Controls and enable AI Pilot.
"The AI gives advice but never proposes actions" Cause: Write Actions is disabled, or your role does not permit write operations. Fix: Go to Settings → AI Agent → Agent Controls and enable Write Actions. If the toggle is greyed out, contact your organization Owner.
"I asked the AI to restart a service but it only explained how" Cause: Ambiguous requests default to Ask mode. The AI recommends rather than acts when uncertain you want it to take action. Fix: Be explicit. Say "Please restart the nginx service on my-server-name" rather than "how do I restart nginx."
"I do not see an AI Pilot menu item in the sidebar" Cause: Your role may be Viewer, which cannot access the AI Pilot interface. Fix: Ask your organization Owner to upgrade your role to Member or Admin.
"Approval card appeared but Allow button is greyed out" Cause: The action is Destructive and requires you to type a confirmation name first. Fix: Type the exact server or site name shown in the red confirmation prompt, then click Allow.
Safety notes
- The AI never performs a destructive action (server deletion, site deletion) without a typed confirmation from you.
- Production Protection prevents write actions on production-tagged servers regardless of your instruction.
- You can deny any approval card at any time. The operation will not proceed.
- All AI actions are logged in the audit trail. See KB-07-13 for details.