Agents' next frontier: agent-to-agent and network effects — Jean-Denis Greze, Town

By ai.engineer

Categories: AI, Tools

Summary

Most AI systems are fundamentally search problems—the goal is getting the right context into the LLM's window before it acts. Jean-Denis Greze (ex-Plaid CTO) reframes agent-to-agent coordination through Coase theorem, revealing why privacy constraints mean true multi-agent systems must strategically approximate access to information rather than grant universal access.

Key Takeaways

  1. Reframe agent architecture as a search problem: success depends entirely on whether the context window contains the right information before tool calls or user results. Better information → better LLM decisions across all context window sizes.
  2. The ideal multi-agent world is actually one agent with access to all world information. This serves as the benchmark test: multi-agent systems should be evaluated on how well they approximate this perfect-information state within privacy/security constraints.
  3. Trust boundary strategy: create agent access matching human team scope (e.g., HR agent with HR-team-level access, security agent across security systems). Popular with IT/security teams but fundamentally limited—doesn't reduce human workload or de-silo data long-term.
  4. Apply Coase theorem to agent design: when all parties have perfect information and zero transaction costs, optimal economic outcomes emerge. Same principle applies to LLM agents—the constraint is privacy/security, not technical architecture.
  5. Evolution of context population: manual (4 years ago) → RAG tools (2 years ago) → agentic search (now). Agentic search gives agents many tools to search content space and find optimal tool calls, but success still hinges on context window preparation.

Related topics

Transcript Excerpt

[music] >> Can you all hear me? All right. Um Well, first things for coming. I can't believe there's anybody in the room, but that's very nice. Uh Uh my name is Jean-Denis. Um I'm CTO at a company called Town. We're not going to really talk about Town, so you can go to town.com and check that out if you want, but that's not the point of the talk today. I was CTO at Plaid for 7 years, and then I was at Dropbox before. And then before that I built software for hedge funds. I've done lots of stuff in my career, and right now I'm working on uh assistance agents for for normal people, not not for engineers, but for like basically everyone in America and the world. And one of the things we've been working on are systems where agents work with other agents. So, agent-to-agent. And the main idea i…

More from ai.engineer