Use a decision people recognize
An agent recalls last year’s price, checks current conditions and recommends what to negotiate. Remembering the price is memory. Selecting relevant information and weighing what to do is cognition.
This definition describes work a system can perform and that you can inspect. It does not assert consciousness or human inner experience.
Guard against both errors, not only one
Calling a model response “thinking” risks treating the system as if it has a point of view it hasn’t earned. That over-attribution is the error this site works hardest to avoid.
The opposite mistake is just as real. Dismissing output as “just processing,” when it selected relevant evidence, weighed competing claims and produced a reason you can check, understates work that actually happened. If it’s all just processing, there’s nothing left to inspect.
Make learning specific
When this site says an agent learns, it means experience can change the durable context, beliefs or processes used later. It does not mean the underlying model is retrained after every interaction.
Show the correction, what changed and the next result. Without that chain, “it learns” is too easy to use as a slogan.
Keep brand language optional
Mind names the persistent context a person or agent holds. The broader thinqOS vocabulary sometimes uses “thinqing.” Neither term should replace an explanation of the mechanism.
Readers should understand the value without adopting our terminology first.
Let evidence set the strength of the claim
A design diagram explains an idea. A working demonstration shows a behavior. A controlled evaluation measures a result under particular conditions. None should masquerade as the others.
Clear language makes the product easier to assess. It also makes the gaps visible, which is necessary if we expect people to trust the work.