← Back to concept explainers

Concept Fable

Error budget

Conclusion first:An Error Budget is the allowed amount of unreliability within a defined period, derived from a service level objective. If the system is within budget, teams may ship changes or take controlled risk; if the budget is burned too quickly, they slow down and focus on reliability. For AI Agent products, error budgets can cover failed runs, unsafe outputs, tool-call errors, latency breaches, human escalations, or customer-visible task failures.

Error Budget · Reliability / operations

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

Fable

The ten cups allowed to break in the ledger

There was a cup workshop in White Porcelain Town that fired thin-rimmed cups for restaurants. The owner initially demanded that "not a single one may break," so the kiln workers would rather take three extra days than try a new glaze color or speed up unloading.

The restaurant owner grew impatient and said that what customers wanted was stable supply, not divine porcelain. Although the cup workshop broke few cups, it often missed banquets, and the restaurant's losses were greater than a few cups.

The owner's first fix was to urge everyone to fire faster. The new kiln's temperature wasn't stable yet, and half a rack of cups shattered in one night. The kiln workers were scolded back into caution, and the shop swung back and forth between "never make a mistake" and "deliver fast."

The old kiln master dug out years of ledgers and found that the restaurant could accept a small number of breakages per thousand cups, as long as restocking was timely and the banquet would not be affected. He suggested writing this acceptable loss into the monthly accounts in advance.

The owner worried this amounted to condoning carelessness. The old kiln master said that allowing ten broken pieces in the accounts did not mean anyone could smash ten at random; it simply told everyone that within the range that did not harm commitments, they could change kiln temperature, test new glazes, and speed up a process once.

The cup workshop set a rule: if broken cups this month approach the upper limit, pause risky changes and focus on stable production; if losses are far below the limit, take a small batch to try a new process. Every broken cup was recorded with its cause, so the same mistake would not recur.

Later, the cup workshop's deliveries became more reliable, and the new glaze was gradually fired successfully. The owner understood: reliability is not never breaking, but knowing how much error the customer can bear, and keeping trial and error and keeping promises on the same ledger.

Most interestingly, the kiln workers became more willing to report broken cups. In the past, a break meant a scolding, so everyone hid small cracks until delivery; now every broken cup entered the same budget account, and they stopped when approaching the limit and only experimented when below it. The cup workshop therefore was neither reckless nor trapped by the slogan of zero errors. The owner later discovered that the ten cups in the ledger let everyone discuss risk rationally for the first time. Fewer breaks was not the only goal, and more experimentation was not willfulness. As long as it stayed within what customer commitments allowed, the workshop could manage stability and progress together. Later the restaurant was also more willing to renew, because it saw that the workshop dared to improve and also knew how to keep its promises. This sentence was later written into the newcomers' handbook, reminding everyone to look at the on-site cost first before talking about saving effort. And precisely because of this reminder, when similar difficulties arose later, everyone would not rush to apply old methods.

Reveal

This story is about: error budgets

An Error Budget is the allowed amount of unreliability within a defined period, derived from a service level objective. If the system is within budget, teams may ship changes or take controlled risk; if the budget is burned too quickly, they slow down and focus on reliability. For AI Agent products, error budgets can cover failed runs, unsafe outputs, tool-call errors, latency breaches, human escalations, or customer-visible task failures.

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

  • Broken cup: user-perceivable errors, failures, or quality problems
  • Ten per month: the error budget derived from the SLO
  • New car, new road: product launches, model changes, and process optimization
  • Stop taking risks when the budget runs out: reliability-first release constraints
  • Public balance: error budget burn rate and observability dashboards
  • The restaurant can accept: customer commitments and business tolerance
  • Final insight: error budgets turn reliability from an absolute slogan into a manageable risk account

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.

SLOSREavailabilityreliabilityincident managementrelease governance

Related distinctions

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

京ICP备2026014272号