Notifications
Clear all
Topic starter
18/08/2026 2:36 pm
TL;DR: Switching LLM providers can fail even when the API call works, because code, prompts, and evaluation history often stay tied to one model, according to Braintrust. The practical lesson is that portability depends on decoupling routing from behavior testing and preserving reusable datasets for future decisions.
NHIMG editorial — based on content published by Braintrust: Avoiding LLM provider lock-in: how to switch models without rewriting your app
Questions worth separating out
Q: How should security teams govern AI model switching without creating lock-in?
A: Treat model switching as a controlled change process, not a simple API substitution.
Q: Why does an OpenAI-compatible gateway not eliminate all provider lock-in?
A: A gateway reduces code coupling, but it does not remove behavioural dependence or data dependency.
Q: What do security teams get wrong about AI model portability?
A: They often assume a working request path means the application is portable.
Practitioner guidance
- Inventory provider-specific dependencies Map every SDK, request format, response parser, fine-tuning job, file API, vector store, and prompt cache that binds the application to one provider.
- Separate routing from evaluation Use a common gateway or abstraction for inference, but keep the acceptance decision in a shared evaluation layer that can score current and candidate models on the same dataset.
- Build a reusable regression corpus Store production traces, labeled examples, metadata, and known failure cases in a provider-neutral dataset so future model reviews start from proven evidence.
What's in the full article
Braintrust's full article covers the operational detail this post intentionally leaves for the source:
- Step-by-step code examples for routing Claude, Gemini, and OpenAI through a shared client interface.
- Practical guidance on comparing current and candidate models with the same dataset and scorers.
- Operational detail on exporting logs, datasets, and experiment results for reuse outside the original provider.
- A worked migration sequence for moving only the traffic that clears acceptance thresholds.
👉 Read Braintrust's guide on avoiding LLM provider lock-in and model migration →
LLM provider lock-in: what changes when you can switch models safely?
Explore further
18/08/2026 3:21 pm
Portable model governance is becoming a prerequisite for safe AI change management. The article shows that model switching fails at three layers at once: code, behaviour, and evaluation evidence. That matters because AI governance is no longer just about model approval, but about whether the surrounding control plane can survive provider change without losing auditability. Practitioners should treat portability as a governance requirement, not a convenience.
A question worth separating out:
Q: How do organisations decide when an LLM is safe enough for production use?
A: They should evaluate the exact workflow, not the model in isolation. Safe enough means the system can verify claims, handle uncertainty, and prevent unreviewed output from changing records, access, or customer outcomes. If those safeguards are missing, the model is not production ready for that use case.
👉 Read our full editorial: Avoiding LLM provider lock-in requires portable code, data, and evals
