Tech Canvas
Technology insights, Authored with Prompts and GenAI
Category: Enterprise AI Platform
A component-by-component teardown of the standard enterprise AI platform stack. Each of the sixteen marketed components — gateways, prompt registries, evaluation platforms, vector databases, agent runtimes, governance tooling — is pushed down to what somebody actually types, configures, or signs, and tested against real work: referral intake, invoice processing, payment variance, care task recommendation.
The finding is that one is genuine infrastructure and belongs to a team you already have. Most of the rest is familiar work under an unfamiliar name.
Written for enterprises that consume AI as a managed service, with a thin slice of custom development. If you train and serve your own models, several of these components are real for you and the argument does not apply.
-
Most AI platform architectures list sixteen components. Taken apart against real enterprise use cases, one is genuine infrastructure — and it belongs to a team you already have.
-
Model gateway, prompt registry, structured output, guardrails. Four boxes on every AI architecture diagram, and for a managed-model consumer they reduce to configuration, source control, one request parameter, and a page of rules.
-
A probabilistic system passes every test your organization runs today and can still be wrong a third of the time. This is the one genuine gap in the AI platform stack, and it is an accountability gap rather than a tooling one.
-
Most enterprise AI arrives inside software you already bought, where none of your internal tooling reaches. Governance has to be expressed as what you require, not what you run.
-
Vector databases, agent runtimes, model registries, feature stores, GPU clusters. Five foundational layers that are either features of software you already run, or infrastructure for a problem you do not have.
-
Almost all of the standard AI platform stack is work your organisation already does under an unfamiliar name. One thing is genuinely missing, and it is not a technology.
Tech Canvas 