Choice: protect tomorrow's options.

Why choice is a survival skill

Every enterprise AI bet right now is being made on top of a stack that did not exist three years ago. Some of those bets will age beautifully. Some will become millstones. Choice is the discipline of building so that, when the model, the vendor, or the regulator changes, your team can pivot in weeks instead of years.

What openness actually buys you

• Open table formats (Iceberg, Delta, Hudi) keep your data readable by more than one engine.
• Open interfaces (SQL, REST, MCP) keep your tools swappable.
• Open semantics — definitions you own, not your vendor — keep your meaning portable.
• Open thinking — patterns that map across clouds — keeps your team employable and your roadmap durable.

Platform independent, platform aware

Choice is not religion. The point is not to refuse every managed service. The point is to keep the parts that are expensive to migrate — your data, your semantics, your evaluation harness — in formats and interfaces you control, while still using the convenience that platforms offer. The book walks through this mapping for Databricks, Snowflake, AWS, Azure, GCP, and the open source side by side.

Designing for change

Build the smallest abstractions that let you swap a model, a vendor, or a region without rewriting the application. The book gives you a checklist — and a sober warning about over-abstracting, which is its own form of lock-in.