Operations
Vendor lock-in
Depending on one AI provider so heavily that leaving becomes impractical.
The AI market moves quickly. Prices change, models are deprecated, capability leadership shifts, providers have outages.
Lock-in creeps in through hardcoded model names, provider-specific features, and prompts tuned to one model's quirks. Reasonable hedges: keep model choice in configuration, prefer open standards like MCP for integrations, and keep your prompts and evals in your own repository so switching is a test run rather than a rebuild.
You do not need to be provider-agnostic about everything. You do need to know what it would cost to move.