The Five Layers of AI Vendor Lock-InDraft

There is a comforting narrative in enterprise AI: "We can always switch models." The API is standardized, the prompts are just text, and if a better model comes along you point your application at the new endpoint and move on. This narrative is dangerously false.

When OpenAI announced the retirement of GPT-4o on March 31, 2026, enterprises discovered what "just switch models" actually means in practice. 57% of IT leaders report spending more than $1 million on platform migrations. And that's within a single vendor's model family — cross-vendor migration is worse.

This whitepaper maps the five distinct layers where AI vendor lock-in occurs — API integration, prompt engineering, fine-tuned models, embedding models, and workflow tooling — explains what breaks at each layer, quantifies the switching costs, and presents the architectural patterns that make model portability an engineering discipline rather than a fantasy. It includes a real case study from our own GPT-4o to GPT-4.1 production migration.

What You'll Learn

• The five-layer model of AI vendor lock-in, from shallow API differences to deep embedding incompatibility
• Why a baseline prompt optimized for GPT-4o scored 4 percentage points worse on GPT-4.1 — and how evaluation-driven migration recovered an 11.5-point improvement
• Why switching embedding models means re-indexing your entire corpus (potentially terabytes of vector data) with no shortcut
• The LLM Gateway pattern and what it solves versus what it doesn't — gateways fix Layer 1 but leave Layers 2-4 untouched
• A concrete migration playbook: audit, evaluate, re-optimize, parallel deploy, document

Who This Is For: CTOs, platform engineers, and technical leaders managing production AI systems who need to plan for model transitions without breaking things.