Uploaded July 2026 | Updated September 2026, 3 weeks ago
As AI agents move from single-agent workflows to multi-agent systems, debugging gets harder. Individual traces can show what happened inside one agent, but they often miss the handoffs, reasoning, retries, approvals, and failures that happen between agents.
In this Observe 2026 session, Ofer Mendelevitch, Head of Developer Relations at Band, explains how agent-to-agent communication works, why multi-agent systems need end-to-end observability, and what changes when agents built in different frameworks need to collaborate across systems, teams, or companies.
Ofer walks through Band’s “agentic mesh” concept: a communication layer for agents that works like chat rooms for AI agents. He also shows why traditional traces are not enough for multi-agent debugging, including clock drift, incomplete handoff visibility, hidden retries, and missing reasoning across agents.
The session closes with a demo of a billing workflow where multiple agents collaborate, introduce an error, and then become debuggable through shared messages, thoughts, tool calls, task events, and end-to-end traces.
Chapters:
00:00 Intro: Ofer Mendelevitch from Band
00:29 Agent-to-agent communication and multi-agent systems
01:09 The future of agent interactions
01:57 Band’s agentic mesh for cross-framework agents
02:42 Chat rooms for agents
03:31 Why single-agent tracing is not enough
04:19 Failure modes in multi-agent systems
05:05 The signals needed for multi-agent observability
05:57 From scattered traces to end-to-end visibility
06:36 Demo: Band agent chat rooms
07:14 Billing workflow with customer support, pricing, and collections agents
08:19 Capturing thoughts, tool calls, and handoffs
09:08 Debugging a multi-agent billing error
10:02 Why end-to-end observability matters
Subscribe to Arize for more talks on AI agents, LLM observability, multi-agent systems, evals, tracing, and production AI engineering.
🔗 Try Arize AX & Phoenix OSS: arize.com
🔔 Subscribe for weekly content on LLMs, agents, and evaluation: youtube.com/@arizeai?sub_confirmation=1
As AI agents move from single-agent workflows to multi-agent systems, debugging gets harder. Individual traces can show what happened inside one agent, but they often miss the handoffs, reasoning, retries, approvals, and failures that happen between agents.
In this Observe 2026 session, Ofer Mendelevitch, Head of Developer Relations at Band, explains how agent-to-agent communication works, why multi-agent systems need end-to-end observability, and what changes when agents built in different frameworks need to collaborate across systems, teams, or companies.
Ofer walks through Band’s “agentic mesh” concept: a communication layer for agents that works like chat rooms for AI agents. He also shows why traditional traces are not enough for multi-agent debugging, including clock drift, incomplete handoff visibility, hidden retries, and missing reasoning across agents.
The session closes with a demo of a billing workflow where multiple agents collaborate, introduce an error, and then become debuggable through shared messages, thoughts, tool calls, task events, and end-to-end traces.
Chapters:
00:00 Intro: Ofer Mendelevitch from Band
00:29 Agent-to-agent communication and multi-agent systems
01:09 The future of agent interactions
01:57 Band’s agentic mesh for cross-framework agents
02:42 Chat rooms for agents
03:31 Why single-agent tracing is not enough
04:19 Failure modes in multi-agent systems
05:05 The signals needed for multi-agent observability
05:57 From scattered traces to end-to-end visibility
06:36 Demo: Band agent chat rooms
07:14 Billing workflow with customer support, pricing, and collections agents
08:19 Capturing thoughts, tool calls, and handoffs
09:08 Debugging a multi-agent billing error
10:02 Why end-to-end observability matters
Subscribe to Arize for more talks on AI agents, LLM observability, multi-agent systems, evals, tracing, and production AI engineering.
🔗 Try Arize AX & Phoenix OSS: arize.com
🔔 Subscribe for weekly content on LLMs, agents, and evaluation: youtube.com/@arizeai?sub_confirmation=1
