Skip to main content
Generative AI has become the new platform layer in technology. The lock-in mechanism is not features or familiarity — it is context: the accumulated instructions, data, tools, and workflows that make a model useful for your specific work.

​
What this piece is

This paper argues that generative AI has become the new platform layer in technology — but the lock-in mechanism is fundamentally different from previous platforms. The binding force is not features or familiarity. It is context: the accumulated instructions, prompt libraries, chat history, indexed data, tool wiring, workflows, and guardrails that make a model useful for your specific work. Originally published on Medium, September 2025.

​
What is the central argument?

Feature gaps between AI providers close fast. Everyone ships function calling, JSON modes, longer context windows, and better reasoning. But your context is unique and sticky. Data gravity, tool gravity, memory gravity, team gravity, and trust gravity create what the paper calls context debt — the hidden cost you carry if you ever need to switch providers. The real platform is the context layer, not the base model or the interface.

​
What are the key concepts?

Context defined. Context is everything the model learns about you and from you that is not the base model weights: system instructions, prompt libraries, chat history, indexed documents, tool schemas, workflows, team artefacts, and governance policies. The four-layer stack. Base model (interchangeable compute), orchestration (routing, agents, evals), context layer (your instructions, memory, tools, data, rules), and interface (chat, SDKs, product UX). In classic computing, users anchored to the interface. In GenAI, users anchor to layer three. Context portability. The paper proposes a Context Transfer Service that would export system prompts, lift chat history with privacy filters, re-embed knowledge bases, translate tool schemas, and replay test suites to validate parity before cutover. Like number portability, but for your AI brain. Practical guidance for builders. Separate context from compute. Abstract providers. Own your knowledge base. Export everything. Design for multi-model. Harden tool contracts. Create an onboarding script for your AI. Measure context health.

​
How does this connect to the wiki’s knowledge areas?

This paper extends the thinking in AI in regulated markets about governance and control, applying it to the platform layer. The context portability argument connects to the trust infrastructure concepts in Open banking and identity — both involve making critical assets portable and user-controlled rather than provider-locked. The practical builder guidance applies directly to Operating models and execution.