Vendor Lock-in
How hard it would be to leave your model provider.
The cost of switching providers: proprietary APIs and features, prompt formats, fine-tuned weights, and integrations built against one vendor’s behaviour.
Practical mitigations: keep an abstraction at the provider boundary, pin model versions, avoid relying on vendor-only prompt features in critical paths, and re-run your eval suite against a second provider occasionally.
The lock-in is usually not the API call, it is the accumulated specificity around it: prompt formats tuned to one model, features only one provider has, evaluation sets you never ported. The cheap insurance is keeping your own interface and your own eval set, so switching is a week of work rather than a project.
The abstraction that paid for itself
A provider retires a model the product is pinned to.
Because requests go through an internal interface and every model is scored against the same eval set, the migration is an afternoon: change a mapping, run the suite, fix the two cases that regressed. Teams without either discover the cost in a firefight.
Standards as the exit route
Integrating with a platform whose tools you want to reuse elsewhere.
Building on an open protocol rather than a vendor-specific tool format means the integration survives the vendor decision. That is the concrete strategic argument for protocol-based tooling rather than feature-specific bindings.
- docsDeprecations ↗
Model and endpoint retirement with notice periods and replacement mappings. The evidence that model pinning is a dependency you have to plan around.
- standardModel Context Protocol specification ↗
The normative document: hosts, clients, servers, JSON-RPC, resources, prompts and tools — plus the security section on consent, user control and care around tool execution.
- blogDonating the Model Context Protocol and establishing the Agentic AI Foundation ↗
MCP moves to vendor-neutral foundation stewardship — the reason it is safe to treat MCP as infrastructure rather than one vendor’s feature.
Model Provider
The company or platform you call to run inference.
Model Routing
Sending each request to the cheapest model that can handle it.
Data Governance
Rules about what data may be used, by whom, and retained how long.
Observability
Being able to see what the system actually did.