Guide

Your AI agent stopped working. What changed?

Start with the dated change and the affected run. If behavior has worsened gradually rather than suddenly, use the agent drift guide for the definition and detection workflow.

Four change-triggered failure modes

This page covers a sudden stopped-working incident. Gradual behavioral decline belongs to the agent-drift guide.

Provider model updates

A provider update can change instruction following, output format, or tool-use behavior even when your prompt file is unchanged.

Impact: A clear change in behavior after the provider update window.

Recent prompt or configuration changes

A prompt patch, tool configuration, policy, or integration change can alter the inputs and constraints the agent sees.

Impact: A regression that starts in the same change window.

Shifting user patterns

A new customer cohort or traffic source can introduce requests the agent was not tuned to handle.

Impact: A concentrated failure pattern in the new request mix.

Stale knowledge and context

Products change, pricing updates, policies shift. The agent's instructions reference a world that no longer exists. Users get outdated or incorrect information delivered confidently.

Impact: Factually incorrect responses that were once accurate.

Start each symptom with one piece of evidence

Do not start by rewriting the prompt. Use the first evidence that distinguishes a sudden incident from a broad behavioral trend.

SymptomFirst evidence to inspect
The agent stops respondingThe affected run, its last successful tool call, and the provider or tool error in the same change window.
Answers change after an updateA before-and-after pair of runs plus the exact model, prompt, tool, or policy configuration used for each.
Only a new request cohort failsThe failing requests alongside the cohort, referrer, or context change that made them different.
Answers become stale or wrongThe affected answer and the product, policy, or knowledge source that changed since it was last correct.

Is this an incident or gradual drift?

If a model, prompt, plugin, policy, or product change lines up with the failure, keep the incident scoped to that run window and inspect the affected evidence before you change anything else.

If the behavior declined slowly with no clear change window, follow the agent drift guide instead of treating a broad pattern as one incident.

Incident triage checklist

First checkNext check
Change windowList model, prompt, tool, policy, and product changesCompare the affected runs before and after the change
Affected behaviorKeep examples of the failing runs and expected behaviorGroup the failure by task, tool, output, or context
Candidate fixTest the smallest change that addresses the evidenceCompare the candidate against the affected and preserved scenarios
After approvalMeasure production evidence and record the outcomeKeep incident evidence distinct from gradual drift

From evidence to an approved change

Preserve the affected evidence

Keep the failing runs, expected behavior, and relevant change window together before editing the prompt.

Identify the narrow failure pattern

Group the failure by task, tool, output, or context. That gives a candidate fix a specific job to do.

Test a candidate fix

Compare a candidate against the affected scenarios and the behavior you need to preserve before approval.

Measure after approval

After an approved deployment, the change can be verified from production data instead of assuming the incident is closed.

Frequently asked questions

My agent worked last week. What changed?

Start with the dated change and the affected run. A provider update can shift instruction following or output format. A prompt, tool, policy, or integration change can alter inputs and constraints. Stale product context can also make once-correct answers wrong.

Why does my agent break after a model update?

Model updates can change how the same prompt is interpreted, including instruction following, output formatting, tone, and tool use. Compare affected runs before and after the provider release before assuming the prompt itself is the cause.

Is this an incident or agent drift?

A sudden failure after a dated change is an incident to investigate from the affected runs. A gradual, broad behavioral decline needs the separate agent-drift definition and detection workflow. Do not collapse the two before you have the evidence.

Investigate the failing run before changing the prompt

Connect your agent to inspect the evidence around a production failure and decide what to test next.

Start for free