AI changes the configuration economics

General-purpose platforms usually contained most of the primitives needed to solve a given business problem. What they did not contain was the person who understood the platforms, configured them, connected them, maintained permissions, built workflows, created interfaces, trained users, troubleshot failures, and kept improving the result.

That labor is the configuration tax, and for a small business it was frequently larger than a decade of subscription fees. Buying preconfigured software was not laziness; it was correct.

The hypothesis is that AI reduces the configuration tax enough to change the answer. Concretely, a business person should be able to state an outcome in ordinary language and have the system determine how to wire the available capabilities together.

A test instruction looks like this: when a potential customer completes this form, create or update the customer record, decide whether the inquiry meets our qualification criteria, notify the right employee, offer available appointment times, create the follow-up tasks, prepare a personalized response, store all correspondence, watch whether the customer replies, and escalate after two business days.

The user describes the outcome. The system decides the wiring. That is the whole transition, and it is either happening or it is not.

Build time from a stated outcome to a working workflow. Expertise required of the person stating it. How much of the result the person can later understand and change. How much maintenance the workflow demands per quarter. How it fails, and whether the failure is visible.

We are also tracking what remains genuinely hard: permission modeling across services, data structures that outlive the first version, long-running processes and their state, error handling that does not silently swallow work, and the absence of a single place to observe a workflow that spans several services.