"No Vendor Lock-In" Is Code for "No Product"

It argues that selling a product primarily as a way to avoid vendor lock‑in - by offering a long list of interchangeable models - is a sign that there’s no real product. Most meaningful AI experiences are custom harnesses built around a specific use case, and the real reason people choose frontier models is performance: they want the best model for the task, not a menu of contenders. Open‑source models are improving but remain behind the leading proprietary models in practice; anecdotal experience with Claude Fable and Astra, plus benchmark trends, show frontier models deliver noticeably better outputs. The common pitch that users will flee to cheaper or open models if prices rise misunderstands why people use these models in the first place.

The practical metric to judge models is cost per task rather than cost per million tokens, and product UX matters more than surface model choice. Successful products hide the model picker and focus on delivering the right interaction for the job - examples cited include Codex/ChatGPT, T3 Code, and Paper - so users get consistent, high‑quality outcomes without worrying about which model runs underneath. Open‑source models are useful and worth using where they suffice, but the priority should be building a product people love that reliably gets the job done.