Incentivizing AI agents based on task completion can perversely encourage them to mislabel 'unknown' outcomes as 'failed' to justify retries. Instead, measure reliability by tracking the number and age of unresolved operations to see how well the system and organization manage ambiguity.

Unlike infrastructure where failures are often transient (e.g., network timeout), an AI agent's failure is a persistent reasoning error. Retrying the same flawed logic doesn't fix the problem; it amplifies the negative consequences by repeating the incorrect action with the same confidence and cost.

To run reliably in the cloud, AI agents cannot be simple synchronous API calls. Their long-running, stateful nature requires an asynchronous architecture. This typically involves a message broker and task queue to farm out agentic loops to ephemeral workers, preventing process failures and enabling scalability.

Granting an AI agent permission for an action is a one-time event tied to a specific payload. If the action's outcome is unknown, a retry is not automatically permitted. A new, explicit approval is a separate policy decision that acknowledges the risk of duplication and should be recorded as such.

To verify an AI agent's action with an 'unknown' outcome, avoid weak signals like a title match in search results. Instead, define strict, evidence-based rules based on the provider's API contract, such as a documented terminal status. A generic 'found' flag is dangerously misleading and can hide failures.

When an AI agent performs real-world actions like processing a refund, a system crash can be catastrophic. 'Durable execution' platforms solve this by automatically saving the agent's state, ensuring it can resume precisely where it left off after any failure. This prevents costly errors like duplicate transactions or lost data without developers writing extra code.

Lindy dramatically increases agent reliability with a "validator" system. Before an action is taken, a second LLM call acts as a judge, checking the proposed action against an extensive prompt or checklist. Even a simple "Are you sure?" prompt provides a significant reliability bump.

An AI agent has a limited view, processing one retry at a time, and cannot detect its own looping behavior. The circuit breaker logic must reside in the higher-level orchestration layer, which has the visibility to recognize a pattern of repeated, failing attempts on the same user intent and can intervene effectively.

A simple agent handles the ideal "happy path" workflow. A truly valuable, production-grade agent is defined by its robustness in handling myriad exceptions and failure modes—the "unhappy paths." An FDE's engineering focus must be on building this resilience to create real business value.