Observability for Tool-Using AI Agents
Learn how to debug AI agents in production with run-scoped traces, stable span attributes, redacted payloads, token and cost metrics, and tail sampling.
Software engineering field notes
Systems, tools, and trade-offs.
Devspedia covers web platforms, cloud systems, AI tools, performance, and security for engineers building and operating software.
Latest
Latest analysis
9 stories on this page
Replayable Test Harnesses for AI Agents
Learn how to test AI agent workflows with recorded traces, deterministic replay, fault injection, and assertions around side effects.
Taming Tail Latency with Hedged Requests
Learn how to cut p99 latency in APIs and AI agents with hedged requests: tuning the hedge delay, hedging only idempotent work, and capping the extra load.
Resource Budgets for Tool-Using AI Agents
Learn how to stop runaway AI agents with token budgets, cost ceilings, step limits, wall-clock deadlines, loop detection, and graceful degradation.
Compensating Actions for Tool-Using AI Agents
Learn how to make AI agent side effects safer with compensating actions, recovery policies, idempotent undo steps, and operator-ready audit trails.
Agent Command Ledgers for Reliable AI Workflows
Learn how to make AI agent side effects recoverable with command ledgers, fenced execution, reconciliation jobs, and replay-safe workflows.
Sandboxing Tool-Using AI Agents
Learn how to run tool-using AI agents behind capability manifests, policy gates, sandboxes, audit logs, and recovery controls.
Designing Durable Agentic Workflows
Learn how to design agentic workflows that survive retries, crashes, tool failures, human approvals, and partial progress without losing intent.
Designing Circuit Breakers for Distributed Services
Learn how to stop cascading failures with circuit breakers that open on real dependency pain, probe recovery safely, and expose clear fallbacks.