← back
Agents' next frontier: agent-to-agent and network effects — Jean-Denis Greze, Town
Takeaway
Agent network effects depend on getting useful information across trust boundaries through acceptable privacy tradeoffs.
Summary
- Greze reframes agent-to-agent systems as a search problem: the decisive model call needs the right information in its context window.
- An imaginary agent with universal data access is used as an ideal benchmark, but privacy and security prevent implementing that design directly.
- Shared agents within household or organizational trust boundaries improve access but create manually maintained silos that do not automatically disappear as models improve.
- Privacy-preserving tools can expose derived signals, such as relationship-strength scores from email, to enable introductions without revealing raw inboxes; designing each tool manually still limits scalability.
agent-to-agentprivacynetwork-effects
Original description
Jean-Denis Greze and his wife share an agent that can read both of their inboxes, including mail from before they were married, and neither of them minds. Greze is CTO of Town and spent seven years as CTO at Plaid. He opens by rejecting his own topic. Agent to agent, he argues, is not a useful concept. Every LLM system is really a search problem: what matters is whether the right information sits in the context window at the moment of the tool call. The ideal is a single agent with access to all the world's information. What blocks it is not context length but privacy, and he reaches for the Coase theorem: perfect information and zero transaction cost produce the optimal outcome, and privacy is that cost. He grades five real strategies by how closely each approximates that impossible agent. A shared trust boundary, such as an HR agent with the access of the team's most junior member, popular today but needing just as many humans as models improve. Custom tools that trade power against privacy, like one that reads everyone's mail and returns only a connection score. Shared silos fed by a sweeper agent that decides daily what may leave a private space, which he thinks pays off soonest. Humans as the conduit, which pings a hundred people to reach one. And a black box that searches every silo unasked, then seeks approval only from whoever owns the information the answer needs. Each leaks in its own way. His own wiki still calls his agent Apex, a month after he renamed it Ivy. Speaker info: https://x.com/jgreze / jeandenisgreze https://greze.com/ Timestamps: 0:00 - Agents that work with other agents 1:20 - Multi agent systems reframed as a search problem 2:30 - The single agent that could see everything 3:24 - The Coase theorem and the privacy limit 4:17 - Strategy one, a shared trust boundary 5:28 - Why that approach does not scale with the models 6:21 - Strategy two, tools that trade power for privacy 8:39 - Strategy three, shared silos and the sweeper agent 11:54 - Strategy four, humans as the conduit 12:49 - Strategy five, the black box 15:31 - Where it breaks, and a wiki poisoned forever 17:32 - Auto mode for privacy, and crossing company lines