Not every model call is Thoughtware

Not every program that calls a model is Thoughtware. The point of the term is to draw a line that is useful to think with, not a purity contest, and not a maturity certificate.

Apply these four questions to any system:

1. Is at least one significant decision made by a cognitive unit, not by code?
2. Is each such cognitive unit named: with declared inputs and outputs, so a person can point at it?
3. Is at least one of the ecosystem layers present in some form, knowledge, experience, retrospection, bridging, or evaluation?
4. Can a cognitive unit be improved on its own: without rewriting the surrounding code?

A yes to all four is a working Thoughtware system. A yes to most, with a clear plan for the rest, is a system on its way.

What does not qualify

Three patterns fail the line even when they use strong models:

• A program that uses a model only to format strings: no real decision is delegated.
• A single monolithic prompt wrapped in a function, the decisions are real, but they cannot be located, shared, or improved separately.
• A chain of model calls with no inputs/outputs contract and no place for feedback to land.

These can become Thoughtware once the decisions are pulled into named units with interfaces and a place for feedback.

Readiness without theatre

It is fine to be embryonic: one cognitive unit doing useful work, little ecosystem yet. It is fine to be anchored: a few units, a clear code boundary, knowledge in some form, meeting real inputs. What is not fine is calling every chatbot “Thoughtware” and then discovering that nobody can say where a decision lives when it fails.

The qualification questions matter more than the label. Use them as a reading, not a brand.