Agents' next frontier: agent-to-agent and network effects — Jean-Denis Greze, Town
Cross-silo agents are primarily a context and privacy-boundary problem. Start with low-sensitivity data, explicit sharing policy, and approval at the moment information leaves a silo.
Greze reframes agent-to-agent work as search: assemble the right context before an answer or tool call. His proposed pattern puts a **sweeper AI** inside each private silo to move policy-approved information into shared company spaces.
For coding-agent systems, define a **low-sensitivity zone** where automated disclosure is acceptable, then require approval for higher-risk sharing. Shared skills and maintained internal wikis can compound useful context across a team without granting every agent raw access to every inbox or system.
Greze reframes agent-to-agent work as search: assemble the right context before an answer or tool call. His proposed pattern puts a **sweeper AI** inside each private silo to move policy-approved information into shared company spaces. For coding-agent systems, define a **low-sensitivity zone** where automated disclosure is acceptable, then require approval for higher-risk sharing. Shared skills and maintained internal wikis can compound useful context across a team without granting every agent raw access to every inbox or system. The policy engine becomes a security boundary. Prompt injection, accidental disclosure, and incorrect facts that persist in a shared wiki remain unresolved; cross-company access requires even stronger trust than the small, high-trust company setting described.
This reframes multi-agent value as governed context propagation rather than more agent-to-agent conversation. Low-sensitivity zones, silo-local sweepers, and approval thresholds offer a path to network effects without universal access, but they make the policy engine and shared wiki durable security and correctness boundaries whose injection, leakage, and stale-fact risks remain unresolved.