Pattern [12]

Exception Handling & Recovery

≈ Try-Catch Blocks / Circuit Breaker Pattern / Retry-Exponential Backoff

> Agentic Definition

Strategies for agents to detect failures (e.g., API errors, hallucinations, logic loops) and self-correct or retry using alternative strategies. This allows the system to be resilient in the face of uncertainty.

> Description

Strategies for agents to detect failures (e.g., API errors, hallucinations, logic loops) and self-correct or retry using alternative strategies. Allows the system to be resilient in the face of uncertainty.

≈ How It Maps to Try-Catch / Circuit Breaker

Preventing system crash upon encountering an error.

≠ Key Divergence

Agentic recovery is semantic. If a tool fails (e.g., "API Error 500"), the agent can reason about why and try a different tool (e.g., "Google Search failed, I will try Wikipedia") or rephrase the query. It is dynamic problem solving, not just retrying the same operation.

> Key Takeaway

Adapt: Error handling is now a creative problem-solving process. You don't just catch the error; you empower the agent to find a workaround.

The Code

Before: Deterministic Error Handling

Deterministic Error Handling
1try:
2    api.call()
3except TimeoutError:
4    time.sleep(2)
5    api.call()  # Retry exact same logic

After: Agentic Recovery

Agentic Recovery
1response = tool.call()
2
3if "error" in response:
4    # Agent reflects on error and pivots strategy
5    plan_b = agent.think(
6        "Tool A failed with error. This path is blocked. "
7        "I should try searching Wikipedia instead to get the data."
8    )
9    plan_b.execute()

Production Notes

  • Without this, agents are brittle toys. Real-world APIs fail; agents must navigate these failures autonomously.
  • Ensure error recovery doesn't spiral into an infinite retry loop of different failed strategies.

Unlock code examples & production notes

Free account — no credit card required.

Sign Up Free

Already have an account? Log in

Frequently Asked Questions

When should I use the Exception Handling & Recovery pattern?

Strategies for agents to detect failures (e.g., API errors, hallucinations, logic loops) and self-correct or retry using alternative strategies. This allows the system to be resilient in the face of uncertainty.

How does Exception Handling & Recovery relate to Try-Catch Blocks / Circuit Breaker Pattern / Retry-Exponential Backoff?

Preventing system crash upon encountering an error. However, there is a key divergence: Agentic recovery is semantic. If a tool fails (e.g., "API Error 500"), the agent can reason about why and try a different tool (e.g., "Google Search failed, I will try Wikipedia") or rephrase the query. It is dynamic problem solving, not just retrying the same operation.

What are the production trade-offs of Exception Handling & Recovery?

Without this, agents are brittle toys. Real-world APIs fail; agents must navigate these failures autonomously. Ensure error recovery doesn't spiral into an infinite retry loop of different failed strategies.