Writer reposted this
Vendor lock-in in the enterprise is a big issue, but generative AI turns the traditional conversation on its head. It’s harder to find a direct equivalent to previous technology waves — ie, choosing SAML over active directory or hybrid cloud vs going all in with a single cloud provider — in generative AI. Why? Because there’s almost no such thing as an LLM-agnostic application — so by definition, you are “locked in” — you can’t build an app where you can easily swap one model for another with no re-work. Continuous rewrites for LLM updates might be manageable for a few apps / use cases, but is hard to scale, and is in fact a reason so few companies have 100s of apps / use cases in production. What enterprises should be optimizing for in generative AI is SELF-RELIANCE, which is the ultimate freedom from vendors. Your proprietary assets as an enterprise in generative AI are the golden four— - Your use cases + business logic - Your data and examples - The in-house talent that can build, iterate, scale, and maintain these apps - The capacity of your organization to learn and change It’s utterly meaningless to have no “vendor lock in” on LLMs — because it all falls down anyway if you change the LLM — and your only path to self-reliance is to have the above. Enterprises should be trying to choose vendors who can truly partner to help build the golden four — partners that don’t look at an enterprise as just another API key. For our part, the Writer full-stack generative AI platform helps companies build highly scalable self-reliance in generative AI by: - Making it easy to separate the model from the business logic - Making it easy to separate the examples and data from both the business logic and the models - Allowing non-engineers to update / change models in a no-code environment - Building LLM families that are “backwards compatible” to minimize rework and not distilling / quantizing models and causing unnecessary rebuild - Making it scalable to manage 100s of generative AI apps and workflows at once