The Escalation Rule: The One Line Every AI Agent Config Is Missing

Source: Dev.to

What Happens Without It

Without an escalation rule, agents in ambiguous situations do their best. That sounds fine until “doing its best” means:

• Making judgment calls it has no business making
• Proceeding on incomplete information
• Creating downstream problems that are hard to trace
• Giving confident‑sounding output that’s quietly wrong

The agent isn’t being malicious; it’s simply following the instruction to complete the task. No one told it what to do when the task can’t be completed cleanly.

The Pattern

The escalation rule has three parts:

1. A Trigger Condition

Define explicitly what counts as “uncertain”, e.g.:

• Missing required input
• Conflicting instructions
• Decisions requiring human judgment (spending money, contacting external parties, deleting data)
• Confidence below a defined threshold

2. A Structured Output

When the trigger fires, write a JSON file (e.g., outbox.json) with enough context for a human to act:

{
  "status": "escalated",
  "reason": "Missing customer ID — cannot proceed with refund processing",
  "context": "Order #4821, amount $89.99, customer email unverified",
  "suggested_action": "Verify customer identity before processing",
  "timestamp": "2026-03-08T07:00:00Z"
}

3. A Hard Stop

The agent writes the file and stops. It does not retry, attempt a workaround, or continue processing until a human intervenes.

In Your Agent Config

Example phrasing in a configuration file (e.g., SOUL.md):

## Escalation Rule
If you encounter a situation where you are uncertain about the correct action,
or where proceeding could cause irreversible harm, STOP immediately.
Write your current state, the reason for stopping, and relevant context
to `outbox.json`. Do not attempt to continue or find a workaround.
Uncertainty is a valid stopping condition.

The last line matters: “Uncertainty is a valid stopping condition.”
LLMs are trained to be helpful; they need explicit permission to not help.

Real‑World Results

Applying this pattern across six agents yielded:

• 80% reduction in downstream errors – most mistakes are caught at the uncertainty point.
• Shorter debugging sessions – outbox.json provides a breadcrumb trail.
• More trust in agent output – when agents proceed, you know they cleared the bar.

The Common Objection

“But this means the agent will stop constantly.”

A correctly configured agent will not. Excessive stops indicate trigger conditions are too broad. Tighten the definition of “uncertain” and stop frequency drops dramatically. An occasional, well‑explained stop is infinitely more useful than a confident, silent mistake.

The 30‑Second Audit

1. Review each agent’s config.
2. Ask: What happens when the agent doesn’t know what to do?
3. If the answer isn’t “it stops and tells me,” you’re missing the escalation rule.

Add the rule today—just one paragraph. Watch your agent operations quality jump.

This pattern and 30+ others live in the Ask Patrick Library — operational configs for AI agent builders. askpatrick.co