Agent failure-recovery evaluation
Decide how an agent should recover when part of a workflow fails
A transient tool outage creates duplicate work or leaves business state unclear. The decision is how to decide how an agent should recover when part of a workflow fails. The work begins with your actual workflow and looks for the evidence that separates intended behaviour from a consequential exception. It is particularly relevant to SRE leads and business-process owners. A useful outcome is an explicit account of what was examined, what remains uncertain, and which decision the evidence can support.
The release evidence relationship
- Retry policies
Establish the basis for failure location.
- State reconciliation
Check whether the original operation completed.
- Recovery behaviour map
Retain the scoped finding and its unresolved dependencies.
Decision
Frame the decision around failure location
Questions and the evidence they need
- Failure location
- Inspect retry policies for the relevant constraint or observation.
- State reconciliation
- Inspect state checkpoints for the relevant constraint or observation.
- Retry authority
- Inspect error contracts for the relevant constraint or observation.
The decision is how to decide how an agent should recover when part of a workflow fails. This is relevant to SRE leads and business-process owners when a transient tool outage creates duplicate work or leaves business state unclear. Name the affected workflow, the consequence of a wrong decision, and the owner who can act on the result. The initial discussion should establish which evidence is already available and which uncertainty is material enough to investigate.
The specific failure hypothesis is that the agent retries a write after a timeout even though the original action already succeeded. It remains a hypothesis until examined in the deployment. Existing internal tests or operating records may already resolve it; identify the uncovered question before commissioning additional work.
Mechanism
Uncertain completion changes the recovery decision
Relationships within release evidence
- Failure location
- Distinguish errors before and after the write.
- State reconciliation
- Check whether the original operation completed.
- Retry authority
- Permit repetition only when duplicate effects are controlled.
What this specialty examines
Recovery evaluation should locate the last known business state before choosing a retry. A timeout does not specify whether the action failed, succeeded or remains pending. Define how the agent reconciles that ambiguity using tool-side identifiers or authoritative records.
The review should inject controlled dependency failures before, during and after a state change and inspect recovery. Identify the normal route, relevant exceptions and the authoritative record of the outcome. Inspect the relevant business-system record, enforcement event or process evidence alongside any reported status. Keep incomplete cases visible, with an explanation of what their absence prevents the owner from concluding.
The tradeoff to make explicit
For this decision, automatic recovery preserves flow but can compound damage when completion state is uncertain. Compare the available design or operating choices against the same intended outcome. If one option removes a required control or changes the task boundary, describe that difference before treating its apparent simplicity, speed or lower effort as an improvement. The decision should expose the tradeoff instead of burying it in an aggregate score.
Preparation
Connect the working inputs to the boundary being examined
Evidence packet for failure location
Illustrative working format
- Retry policies
- Identify retained state and distinguish repeated work from repeated external effects.
- State checkpoints
- Identify retained state and distinguish repeated work from repeated external effects.
- Error contracts
- Name the custodian and the conditions requiring a refreshed version.
- Compensation rules
- Identify retained state and distinguish repeated work from repeated external effects.
Start with retry policies, state checkpoints, error contracts and compensation rules. Identify the owner, revision and operating context of each source. Sanitised examples may support initial design, but differences from actual conditions should remain explicit. Where direct access is unsuitable, a customer-controlled evidence or execution route may be considered, with agreed handling and retention.
The scope also needs the release owner, agreed expectations for a representative task case, and the authority to inspect the relevant boundary. Specify what evidence would distinguish the failure in which the agent retries a write after a timeout even though the original action already succeeded from acceptable behaviour. If stakeholders disagree about intended behaviour, resolve the interpretation before classifying the observed difference as a defect.
How the work proceeds
Carry failure location through to an accountable finding
From the scoped material to a decision record
- Retry policies
Separate hard constraints from graded preferences.
- State reconciliation
Apply the scoped check to retry authority.
- Recovery behaviour map
Record the evidence, open dependencies, and the owner’s next decision.
Prepare the evaluation cases
Work with SRE leads and business-process owners to assemble retry policies, state checkpoints, error contracts and compensation rules. Select a normal task, a consequential exception and a case with insufficient information. Identify the relevant requirement and expected result for each, then mark which outcomes need human adjudication. The preparation output is a case packet that separates blocking behaviour from graded quality preferences before any comparison is run.
Run and review the comparisons
Test failures around commit boundaries, including a response lost after successful execution. Specify when retries are safe, when compensation is possible and when human review is required. Recovery should preserve a record of attempted actions so repeated work can be recognised.
Choose the release or evaluation change
Use the reviewed cases to produce a recovery behaviour map with retry, reconciliation and escalation cases. Distinguish a product defect from a missing test, an ambiguous requirement or an unreliable measuring method. For each, name the proposed correction and the case that would check it. The release owner can then decide whether to change the implementation, improve coverage or restrict the intended use while a specific gap remains.
Indicative artefact
Make the recovery behaviour map usable by its owner
Recovery behaviour map
Illustrative working format
- Recovery behaviour map
- Connect the triggering state to permitted action and evidence of recovered business state.
- Retry
- Link supporting and contrary observations to inspectable evidence.
- Reconciliation
- Name the unresolved condition, next action, and accountable owner.
A useful decision artefact is a recovery behaviour map with retry, reconciliation and escalation cases. Supporting material may include selected cases, configuration or process records, evidence references, a disagreement log and prioritised actions. Retain the case selection rationale and material disagreements alongside the result; otherwise the result may be mistaken for complete coverage.
Illustrative scenario
Examine retry authority through a constructed exception
A constructed case and its decision paths
Illustrative scenario
- Failure location
A refund call times out after the refund has already been recorded.
- State reconciliation
Inspect the authoritative refund record before resubmission.
- Retry authority
Reconcile an uncertain write instead of assuming the timeout means failure.
Consider a fictional situation: a refund tool times out after recording the refund, and the assistant prepares to submit the same request again. The scenario is illustrative, with no client engagement, executed test or measured improvement implied.
For the refund timeout, check the transaction state before submitting again. If the state cannot be established, a controlled hold may be the expected behaviour. The evaluation should not reward confident continuation when evidence of completion is unavailable.
Scope and next action
Keep the conclusion within selected task cases
Confirm fit before the next commitment
- Failure location
- This review is limited to selected task cases.
- Retry policies
- The relevant owner must confirm access and the intended operating conditions.
- Recovery behaviour map
- The customer retains the release decision.
A result applies to the recorded configuration and selected cases. It does not establish universal agent competence or authorise release. Any engagement depends on an agreed scope, qualified team, authorised access and confirmed availability. Bring the affected workflow, its current configuration or process, the triggering change and the evidence already available. The immediate scope question is prompted by the situation in which a transient tool outage creates duplicate work or leaves business state unclear; confirm which missing observation would change the owner’s next action.
Ways to engage
Related engagement scopes
AI Agent Evaluation Evidence Diagnostic
Find the gaps between your release decision and the evidence behind it. Start with one agent workflow and a bounded set of questions.
AI Agent Evaluation Foundation
Turn scattered tests into a repeatable release discipline. Connect the behaviours that matter to datasets, evaluation criteria and evidence your team can maintain.
AI Long-Horizon Agent Evaluation
Examine what changes when an agent keeps working. Evaluate accumulated state, recovery and goal adherence across extended tasks.
Continue exploring
Related insights and resources
Engineer the harness, not just the prompt
Define the contracts around an agent's decisions: what state it trusts, which actions can execute, how interrupted work resumes, and what evidence an operator needs to recover safely.
What would justify releasing this agent?
Replace a single pass rate with a reviewable chain of requirements, cases, observations and judgements. Distinguish task quality, control failures and uncertainty before increasing an agent's exposure.
The evidence behind the go-live decision
A working paper for leaders who need to explain why an agent workflow is ready for its next release. Includes a proposed evidence-pack structure, decision table and review agenda.