Why Your AI Agents Can't Talk to Each Other (Yet) — Vlad Luzin, BAND
- Source
- youtube.com
- Author
- AI Engineer
- Date
Context
Vlad Luzin, co-founder and CTO of BAND, argues that people now act as the router between AI agents, copying output from one session to another, and that infrastructure should do this instead. In his AI Engineer World's Fair 2026 talk, he says a single agent suffers from confirmation bias and dilution however large its , and ties that to using adversarial agents, such as a reviewing session that checks a planning session.
He says chat apps and agent protocols such as (which connects agents to tools) and A2A (which connects agents to each other) do not solve this, because connecting agents is a distributed-systems problem needing ordered real-time transport, , runtime binding, identity and observability. The talk ends with the launch of BAND's interaction layer and demos of Claude Code, Codex and LangGraph agents working together. The vendor selling that layer is the source of the demos and the claimed protocol gaps, so these are claims rather than comparisons.
- attention — The mechanism that lets a model weigh which earlier words matter for the word it's currently processing — the core operation of a transformer.
- Context and context window — Context is the information available for a generation; the context window is the capacity available to hold it.
- Model Context Protocol (MCP) — A published agreement for an AI application to discover what a separate program offers and ask it to run one of those offerings.
- Persistence — The property of surviving beyond the process that produced the information, usually by being written to a file.
videoYour Agents Are in Solitary Confinement: Why MCP & A2A Aren't Enough — Vlad Luzin, BandAI Engineer
videoHow to Get Your Org to Adopt Coding Agents (Without Shipping Garbage) — Eyal Blum, FigmaAI Engineer
videoDecoupling the Brain from the Hands: Claude Managed Agents (Part 1 of 4) — AnthropicAI Engineer
Checking sign-in…
Loading comments…



