How Do You Know What Your Agent Is Actually Doing?
Last Updated on August 24, 2026 by Editorial Team
Author(s): Nitin Bisht
Originally published on Towards AI.
How Do You Know What Your Agent Is Actually Doing?
A few weeks ago, I watched an AI agent burn through 40,000 tokens calling the same search tool six times.
The article argues that conventional “no error” monitoring is insufficient for AI agents because agents are non-deterministic, can fail while still producing outputs that look successful, and can turn a single user request into many nested model/tool calls that drive cost and latency. It proposes treating an agent run like a tree of events and capturing observability data (inputs/outputs, timing, token usage, costs, tool choice, and context) using tracing—ideally via OpenTelemetry—to make the full decision path visible. It also stresses separating tracing (what happened) from evaluation (whether it was good), using methods like LLM-as-a-judge, golden datasets, and human review. Finally, it highlights common failure patterns (tool loops, retrieval failures, behavior changes, and cost spikes), notes you don’t need a massive stack to start, and concludes that the real point of agent observability is enabling answers to “why did the agent do that?” rather than ending up with “the logs say it succeeded.”
Read the full blog for free on Medium.
Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.
Published via Towards AI
Towards AI Academy
We Build Enterprise-Grade AI. We'll Teach You to Master It Too.
15 engineers. 100,000+ students. Towards AI Academy teaches what actually survives production.
Start free — no commitment:
→ 6-Day Agentic AI Engineering Email Guide — one practical lesson per day
→ Agents Architecture Cheatsheet — 3 years of architecture decisions in 6 pages
Our courses:
→ AI Engineering Certification — 90+ lessons from project selection to deployed product. The most comprehensive practical LLM course out there.
→ Agent Engineering Course — Hands on with production agent architectures, memory, routing, and eval frameworks — built from real enterprise engagements.
→ AI for Work — Understand, evaluate, and apply AI for complex work tasks.
Note: Article content contains the views of the contributing authors and not Towards AI.
Related posts
Recent Posts
