Agents' next frontier: agent-to-agent and network effects — Jean-Denis Greze, Town
Summary
The talk reframes multi-agent systems not as 'agent-to-agent' interactions, but fundamentally as a search problem: ensuring that an LLM's context window contains the optimal information for a tool call. The primary technical barrier to achieving this ideal state—a single agent with access to all world information—is not context length, but privacy and security. Greze outlines five strategies (Shared Trust Boundaries, Custom Tools, Shared Silos, Human Conduit, Black Box) that attempt to approximate the optimal outcome while managing data leakage risks.
Key takeaways
-
Reframing Agents as Search Problems
2:00
Most LLM systems are best viewed as search problems. The goal is engineering the system so that the context window contains the precise information needed before a tool call, maximizing the LLM's ability to return the best result.
-
The Privacy Constraint (Coase Theorem)
5:24
The ideal state is one agent with access to all world information. However, privacy acts as a transaction cost, preventing this perfect data aggregation, which the Coase theorem highlights.
-
Shared Silos and Sweeper Agents
13:59
A promising approach is creating shared silos (e.g., wikis) where a 'sweeper AI' automatically identifies and moves information from private spaces into public, company-wide knowledge bases.
-
The Black Box Approach
This advanced method allows an LLM (in a 'black box' agent) to query multiple silos without pinging every human. It only asks the data owners for approval at the final moment of information sharing.
-
The Future: Auto Mode
The frontier is 'auto mode,' where LLMs are trusted to automatically determine if a disclosure is low-sensitivity and can be shared without explicit human approval, scaling with model capacity.
Technical details
-
Agent Architecture Reframing
120s
The concept of 'agent-to-agent' is discouraged. The focus should be on optimizing the context window for a single, powerful agent call.
-
Coase Theorem Application
324s
In data systems, the theorem suggests that perfect information and zero transaction cost yield optimal outcomes. Privacy represents this non-zero transaction cost.
-
Strategy: Shared Trust Boundary
417s
Giving an agent access to multiple, related systems (e.g., all HR systems) as if it were the lowest-level employee in that team. Limitation: This approach does not scale with model improvements and still requires human thought.
-
Strategy: Custom Tools (Power vs. Privacy)
621s
Building tools that perform specific, limited data aggregations to trade power for privacy. Example: A tool that reads all emails but only returns a 'relationship strength score' between two parties.
-
Strategy: Sweeper AI
839s
An AI agent placed within private silos with a policy to automatically detect and surface information into shared, public spaces (e.g., company wikis). This is the most promising approach for internal data flow.
-
Black Box Approach
A system where an LLM has full access to all silos but only requests human approval when it must make a specific, actionable disclosure (the 'sharing moment').
Mentioned resources
- Town
- Apex / Ivy
Channel & topics
Watch on YouTube · Back to latest
This independent, AI-assisted summary is provided for commentary and informational purposes. It may contain errors or omit important context. Please watch the original video for the creator's complete presentation. Video, thumbnail, and related copyrights belong to their respective owners.