RELIABILITY AND ERROR RECOVERY

CHECK YOUR UNDERSTANDING • Why is retrying immediately and endlessly harmful? ◦ Answer: It adds load to a struggling service, wastes money, and can run forever. • Which errors should not be retried? ◦ Answer: Permanent errors such as permission denied or bad request, which will fail every time. • Why must loop limits be enforced by the harness? ◦ Answer: A looping model may ignore its own instructions, but the harness code reliably stops the run. • What must an agent do when it uses a backup source? ◦ Answer: Tell the user, so they are not misled. • How do you decide how many retries a system deserves? ◦ Answer: Weigh the cost of failure, how often it occurs, and what each protection costs, so high-stakes systems get more protection. PART FOURTEEN: DEPLOYMENT, OPERATIONS, AND AGENT OPERATIONS (AGENTOPS) Building an agent that works on your laptop is one achievement. Keeping that agent working for thousands of real people, day after day, is a different and harder one. Part Fourteen is about the second achievement. Deployment means putting software into the place where real users can use it. Operations means the everyday work of keeping it healthy. Agent Operations (AgentOps) is the name for operating agents specifically. We begin with the journey from a prototype to a product.