Beyond Logs: Debugging the Black Box of Multi-Step AI Agents!
AI agents don't behave like traditional software. The same input can lead to different tool calls, reasoning paths, and failure modes, making them difficult to debug and monitor in production.
Join us for a VickyBytes Meetup with Chayan Deokar, OpenShift Engineer at Red Hat and Technical Speaker, as he takes us beyond traditional logs and explores how to build observability into multi-step AI agents.
In this session, we'll explore how to:
Observe the plan → act → observe loop
Trace multi-step calls across tools and sub-agents
Identify silent failures where an agent appears successful but produces the wrong outcome
Attribute cost, latency, and reliability issues to the exact step causing them
Build a practical observability checklist for AI agents
The session will also include a Q&A, giving attendees a chance to discuss their own challenges around AI agent observability.
What you'll learn
By the end of the session, you'll have:
A practical mental model for AI agent observability
A checklist for instrumenting multi-step agents
Better ways to trace tool and sub-agent interactions
Approaches to identify silent failures
A clearer understanding of cost, latency, and reliability across agent workflows
Who should join?
Developers, AI/ML engineers, DevOps engineers, and anyone building or working with AI agent systems.
Format: Online via Google Meet Date: 27th August 2026 & 5 PM
Come learn how to treat observability as a core part of agent architecture, not an afterthought.
Speaker intro
Chayan Deokar is an OpenShift Engineer at Red Hat and a technical speaker. He is a Certified GCP Solution Architect and holds three Red Hat certifications. He shares his expertise through technical talks and discussions with the developer community.
Connect with speaker: chayandeokar