Agentic Observability for AI Agents

di Srinivas Bommena (Autore)

Srini, 2026

AI agents need more than monitoring. They need an operating model for evidence, evaluation, reliability, economics, and governed action. Agentic Observability for AI Agents develops a new way to operate these systems. The book moves beyond conventional logs, metrics, traces, and dashboards to introduce a runtime evidence architecture that connects what an AI agent did with whether the behavior was correct, reliable, economical, safe, and authorized. Its central progression is from observability of AI agents to agentic observability. Rather than treating telemetry as something humans inspect only after an incident, the book develops an operating loop in which runtime evidence supports investigation, evaluation, recommendation, governed authority, action, and continuous learning: Evidence ? Reasoning ? Recommendation ? Authority ? Action ? New Evidence At the center of that model is the Observability Operations Agent—a bounded operational agent that can correlate evidence, reconstruct executions, compare hypotheses, request approved diagnostics, and recommend next actions while remaining inside explicit policy, runbook, permission, and approval boundaries. Inside the book, you will learn how to: recognize silent AI failures where infrastructure is healthy but the business outcome is wrong; assess observability maturity by what a production system can actually explain and prove; use OpenTelemetry as a portable evidence substrate for GenAI systems; model LLM operations with the evidence needed for diagnosis, correlation, and evaluation; trace RAG pipelines through retrieval, reranking, provenance, context assembly, and generation; reconstruct complete agent trajectories, including tools, MCP interactions, delegation, handoffs, retries, memory, and state; observe multimodal pipelines where different modalities create different evidence boundaries; balance sampling, privacy, retention, and telemetry cost; design evaluation systems using deterministic checks, model-graded evaluation, human judgment, golden datasets, and trajectory evaluation; connect evaluation directly to runtime telemetry so quality judgments are tied to the exact execution that produced them; use replay, shadow evaluation, regression suites, canaries, and CI/CD quality gates for defensible release decisions; detect hallucination, drift, degradation, changing behavior, and long-tail failure patterns; observe safety controls, guardrails, abuse signals, prompt injection, and excessive agency; attribute token, model, retrieval, tool, retry, workflow, and human-review costs to real tasks and business outcomes; engineer latency and performance across complete agent journeys rather than isolated model calls; debug AI incidents through evidence chains, trajectory comparison, replay, and hypothesis testing; turn production failures into new evaluation cases, regression controls, and a governed feedback loop. Instead of switching between disconnected toy examples, the book follows AsterVale Group, a recurring fictional travel, mobility, and service enterprise. An optional companion package provides practitioner activities, synthetic evidence, traces, scorecard

• Autore: Srinivas Bommena
• Anno edizione: 2026
• Editore: Srini
• Formato: EPUB3 con DRMFREE
• Lingua: Inglese
• Compatibilità: Tutti i dispositivi (eccetto Kindle)
• Layout: Reflowable
• EAN: 9798237891133