← Back to concept explainers

Concept Fable

Upgrade path

Conclusion first:An Escalation Path is a predefined route for reporting and handing off exceptions. When AI / Agents encounter high-risk, low-confidence, insufficient-permission, compliance-sensitive, customer-complaint or undeterminable situations, the system should know which person, role, queue or process to hand the issue to. A clear escalation path prevents the Agent from continuing to guess, continuing to err or failing silently.

Escalation Path · Operations

Read the story first. There is no rush to give a definition here; let the question grow on its own first.

Fable

The red card in the little postman's hand that says "find whom"

The postman of Qingshi City didn't just deliver letters—he also handled small troubles. If an address was wrong, he'd ask the neighbors; if the recipient had moved, he'd check old registrations; if the envelope was torn, he'd reseal it. Most problems, he could solve on the road.

City residents thus grew increasingly dependent on the courier. But one day he received a letter stamped with a court seal, with a blurry address and a recipient name similar to two people. Delivering it wrong could affect a lawsuit; not delivering it would miss a deadline.

He carried the letter around the streets for three laps, asked two neighbors, and checked the old registry. Everyone gave him a little clue, but no one could take responsibility. In the end he put the letter back at the post office and said he would deal with it tomorrow. The next day, the deadline had passed.

The postmaster did not blame the courier. He asked: did you know who to go to at the time? The courier shook his head. He knew he could not solve it himself, but did not know where the problem should go.

So the post office made a red card and issued it to every courier. The red card listed several situations: if legal documents were uncertain, find the clerk; if there was a monetary dispute, find the head bookkeeper; if the recipient's identity conflicted, find the household registration officer; if someone complained of threats, find the duty supervisor. If it could not be determined within half an hour, escalate first; do not keep guessing.

The red card is not meant to make the postman do less, but to stop him from relying on nerve alone when something exceeds his authority and judgment. Ordinary letters are still handled by the postman; special letters have a clear destination; emergencies are taken over by the duty officer.

A few weeks later, another similar court letter arrived. The postman saw the conflict and immediately handed it to the clerk according to the red card. The clerk checked the court register, delivered it to the right person the same day, and left a handling record.

The courier remained the fastest person in the city. The difference was that he knew which stretch of road he should run himself and which stretch had to be handed to someone who could take responsibility. The little red card protected not just the courier but the whole city's judgment. Later, when the commissioner inspected, he checked not only whether the couriers had delivered the letters but also whether they had handed things over promptly when they should. Holding on and delaying looks like responsibility but actually hides risks you can't bear yourself. The red card lets responsibility move along the right path. Later, when new couriers joined, their first lesson wasn't memorizing every street in the city but recognizing the moment when they should stop guessing. Knowing when to stop and find the right person is also a skill.

Reveal

This story is about: upgrade paths

An Escalation Path is a predefined route for reporting and handing off exceptions. When AI / Agents encounter high-risk, low-confidence, insufficient-permission, compliance-sensitive, customer-complaint or undeterminable situations, the system should know which person, role, queue or process to hand the issue to. A clear escalation path prevents the Agent from continuing to guess, continuing to err or failing silently.

What matters is that it is not just a term, but a structural problem that repeatedly appears in real AI / Agent systems. Understanding it lets you judge when to add a model, when to change a process, and when to add governance.

Metaphor mapping

  • Courier: frontline Agents or automated workflows that can handle large volumes of routine tasks
  • Letters with a court seal: high-risk or compliance-sensitive tasks
  • Ambiguous addresses and recipients with the same name: low confidence, identity conflicts or insufficient context
  • Circling the streets three times: the system repeatedly tries but never truly solves the problem
  • Not knowing who to ask: missing escalation paths
  • Red card: a predefined Escalation Path that specifies exception categories and recipients
  • Clerk, accounting supervisor, household registration officer, shift leader: the responsible person or processing queue for different anomalies
  • Escalate if no decision can be made within half an hour: trigger conditions to avoid endless delay or blind execution
  • Give uncertainty an exit: escalation paths let Agent systems safely hand off to humans or specialized processes at the boundary

Andy Ren Xu's judgment

The practical value of this concept is that it helps you break "AI features that look smart" into deliverable, verifiable, and governable units of work.

human oversightapproval gateincident managementexception handling
京ICP备2026014272号