← Back to concept explainers

Concept Fable

Autonomy level

Conclusion first:Autonomy level is a grading of how independently an Agent works: it can only recommend, draft first, execute after approval, execute autonomously within limits, or run fully automatically in specific scenarios. In enterprise governance, the autonomy level determines the mode of human intervention, approval thresholds, audit requirements, and risk boundaries. If Andy Ren Xu is to deliver service results, he cannot vaguely promise "AI completes it automatically," but must explain which tasks can be automated, which require human confirmation, and which must be escalated.

Autonomy Level · Agent Operating Model

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

Fable

Before the apprentice gets the key, he first gets a tiered door plate

In South Town there was a repair shop. The apprentice started by sweeping the floor, then gradually learned to measure wood, cut boards, and patch beams. At first the shopkeeper thought skill was just skill—whoever was fastest got more keys.

There was an apprentice who repaired doors quickly, so the shopkeeper sent him alone to fix the ancestral hall roof. He could replace planks, but he did not know that an old beam was pressing beneath the roof. After a few hammer blows, rainwater ran through the cracks into the ancestral tablet room. Quick craftsmanship turned into a big problem.

The shopkeeper hurriedly took back all the keys and ruled that any job must be accompanied by a master. The shop was safe, but painfully slow. Even changing a pane of window paper required a master's nod, and customers complained that the repair shop felt like a government office.

The old carpenter suggested layering the work. The first layer allowed apprentices only to hand over materials and keep records; the second layer could repair simple items according to a checklist; the third layer could handle common faults alone; and only above that was judgment of structure, changes to plans and adjustment of staffing permitted.

They also gave each layer a fallback. When apprentices encountered unfamiliar beams, cost overruns, or sudden changes from customers, they had to stop and ask; for routine minor repairs, there was no need to call the shopkeeper back from his meal.

At first, the young workers grumbled that there were too many door plates. It wasn't until the same roof leak happened that they realized: one person only needed to replace tiles, one had to judge the load-bearing structure, and one had to arrange for the whole courtyard to be shut down. Only then did they see how far apart "can do a bit" and "can be responsible for a section" really are.

The repair shop later became fast again, but no longer judged everyone with a single 'can or can't.' The shopkeeper understood that autonomy isn't handing over a master key, but delegating authority level by level according to risk, scope and conditions for asking back.

The doorplate was later posted on every toolbox and written into customer orders. Customers knew who would come and how far the work could go; apprentices knew when they should continue and when they should stop. The shop had far fewer vague promises, because “leave it to him” was no longer just a phrase, but a clearly graded responsibility. Later, when the shopkeeper evaluated apprentices, he did not look only at craftsmanship, but at whether they knew their own level. Being able to proactively complete the work that should be completed, and promptly asking when crossing a boundary, was what counted as maturity. The clearer the autonomy level, the less responsibility could be shifted after an incident. They also knew that escalation was not an award title, but proof that one could hold a larger scope. This sentence was later written into the newcomers' handbook, reminding everyone to look first at the cost on site before talking about saving effort. And because of this reminder, when similar difficult problems arose later, everyone would not rush to apply old methods.

Reveal

This story is about: autonomy levels

Autonomy level is a grading of how independently an Agent works: it can only recommend, draft first, execute after approval, execute autonomously within limits, or run fully automatically in specific scenarios. In enterprise governance, the autonomy level determines the mode of human intervention, approval thresholds, audit requirements, and risk boundaries. If Andy Ren Xu is to deliver service results, he cannot vaguely promise "AI completes it automatically," but must explain which tasks can be automated, which require human confirmation, and which must be escalated.

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

  • The old shopkeeper holds all decisions alone: fully manual execution
  • Give all keys to the apprentice: excessive automation without tiering
  • Approving everything: putting a human in the loop even for low-risk work slows it down
  • Five-color door plates: different Autonomy Levels
  • White, blue, yellow, red, black badges: permission tiers from automatic handling to prohibited execution
  • Amounts out of bounds and unclear identity are automatically sent for review: autonomy levels are tied to escalation conditions
  • The apprentice gradually earns more yellow cards: autonomy levels should be calibrated to performance and risk

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-in-the-loophuman-on-the-loopapproval gateagent operating model

Related distinctions

This concept is easily confused with adjacent concepts; it is recommended to continue to the comparison page.

京ICP备2026014272号