DP7 - Simplicity and Interoperability

Gaowei Chang sent these notes on DP7 by email:

Thank you for inviting me to review DP7. I strongly agree with the chapter's central idea: interoperability is not just about API or data format compatibility, but also about preserving identity, meaning, and relevant rights across system boundaries. Based on some of our experience researching agent protocols, I have a few suggestions for your consideration.

1. Distinguish cross-system interaction from portability The chapter spends considerable time discussing how identities, data, reputation, and other assets can move between systems. However, I think there is another equally important dimension of interoperability: enabling different systems to interact directly without requiring identity or data migration. For example, an agent should be able to maintain its own identity while discovering and interacting with agents or services on other platforms. I suggest treating cross-system interaction and cross-system portability as two relatively independent interoperability goals.
2. Separate identity, trust, and authorization I strongly agree with the discussion of identity continuity in Section 5.7. In our research, we have also found that cross-platform identity authentication is an important foundation for agent interoperability. However, identity verification does not automatically establish trust or authorization. Even if an agent uses the same identity across multiple services, each service still needs to independently determine what that agent is allowed to do. This distinction could be made more explicit in the sections on identity continuity and authorization. Similarly, permissions or governance policies from one system should not be assumed to be automatically accepted by another.
3. Give more attention to simplicity The chapter provides a thorough discussion of interoperability, but simplicity seems to receive relatively less attention. In our experience, reusing HTTP, existing Web infrastructure, and established standards, while separating core capabilities from optional extensions, can significantly reduce implementation complexity. Perhaps the chapter could introduce a minimal interoperability baseline, along with a few practical evaluation criteria. For example, how much additional work would two independently developed systems need to do before they could achieve basic interoperability?
4. Consider dynamic protocol negotiation between agents This is an interesting direction we have been exploring recently. Traditionally, interactions between Web systems usually rely on predefined interfaces and data formats. Although protocol negotiation mechanisms already exist, AI agents introduce a new possibility: two agents may be able to dynamically negotiate which protocols, interfaces, and interaction methods to use based on their capabilities, task goals, and constraints. For example, two agents could first exchange capability descriptions and then decide whether to use a structured API, message-based interaction, or another mutually supported approach. When necessary, they could also negotiate specific data formats and security requirements. This means future interoperability may not depend entirely on developers defining every interaction detail in advance. Agents themselves could participate in establishing how they communicate, while still operating within clearly defined security and authorization boundaries. I think this is an important difference between the emerging agent internet and traditional Web interactions. It may be worth considering whether this deserves some discussion in the chapter.
5. Add a concrete agent interoperability example I suggest including a simple cross-platform scenario: a personal agent discovers a service operated by another organization, reads its capability description, verifies its identity, obtains the necessary authorization, and then interacts with it directly. Combined with the dynamic protocol negotiation discussed above, this could provide a concrete illustration of how agents might achieve interoperability without depending on a shared centralized platform.
6. Balance auditability with privacy The interoperability receipts and provenance mechanisms proposed in the chapter are valuable. However, requiring verifiable records for every cross-platform interaction could expose sensitive information or allow different platforms to correlate a user's activities. I suggest making audit requirements proportional to the risk of the operation, with clearer consideration of what information needs to be recorded, who can access it, and how privacy is protected. Thank you again for the invitation. I hope these suggestions are helpful as you continue refining the chapter.