Two Principles: Feed Context, Check the Work
Two rules govern every interaction with the tool. They thread through everything else.
Good use: feed it your notes, transcripts, data, documents. Ask it to synthesize, summarize, reorganize, extract.
Bad use: asking it to invent novel content from nothing, or trusting it to know your situation when you haven't told it.
I once tried building a custom assistant loaded with Savory materials. It performed worse than the plain model — circular and non-committal. More context doesn't help if the context isn't right.
You would never take a transcribed doctor's visit and let AI auto-send the prescriptions. The same logic applies to land management.
The rule: every AI workflow includes a step where a knowledgeable human reviews the output before acting on it.
Pair them: feed it, then check it. That's the loop.