We need to distinguish what justifies AI to become involved in certain work practices and job roles for the purpose of improving quality services, products and efficiency. It must be engraved in our mindset that AI is a double-edged sword that, if applied sluggishly and without much foresight, is not going to improve productivity or quality and could in fact have the reverse desired effect.
The avoid-list above is only half the picture. The positive rule, agreed during
the first agent configuration pass:
(Recorded examples: the chat-Q&A anti-pattern and its consequences are
documented on the Smart Scheduling plan
§7, together with the product decisions of 2026-08-20/21.)
Before any AI feature is proposed or approved, it must answer these five
questions:
A "no" on questions 1–4, or a "no" on 5, is a strong signal to stop and
reconsider.
There is a right and a wrong way to use AI, and it applies to us as much as to
the product.
The wrong way is to hand the thinking over entirely: raise a problem and let
the AI come up with everything. That is the laziness the avoid-list warns
about — it produces complacency and, sooner or later, worse outcomes.
The right way is working WITH AI: the professional makes their own
analysis, exercises judgement, and puts their understanding into words; the AI
then extends, structures, challenges and elaborates on that clear picture with
more depth. The human steers; the AI assists.
A worked example (2026-08-21): the Smart Scheduling discussion. The user
raised an issue about how AI should be used in the front-desk workflow, then —
instead of expecting the AI to come up with the answers — made their own
analysis and put it into words. That gave the AI a very clear picture, which
it was able to expand on more elaborately and with more depth. The result was
a stronger standard than either could have produced alone.
This is exactly what "assist the user in making sure they are making the right
decision based on the facts and relative circumstances" looks like in
practice: the AI's value is amplified when the human has already done the
analytical work of framing the problem.
Every proposed AI feature must answer the justification checklist in writing,
and must be judged against both the positive standard and the avoid-list,
before it is added to any plan. The checklist is deliberately conservative:
when in doubt, the default is that AI is not justified.