Raju Dandigam

The “treat agents like software” point is exactly the right framing. Once an agent has tools, memory, and external side effects, the hard part becomes contracts, traces, and failure recovery. I’d especially emphasize replayable execution logs because standard stack traces do not explain why an agent chose a path. Production readiness should include the ability to inspect every model call, tool call, and state transition after the fact.

The line between DevOps and AgentOps is becoming very real. Traditional DevOps assumes behavior is mostly determined by code and config, but agent behavior also depends on prompts, retrieval, memory, tools, and model routing. That means observability needs to capture intent and decision paths, not just latency and errors. I’d love to see AgentOps practices standardize around trace schemas for agent steps and tool calls.

Great article!