Guides
What an assistant can and cannot responsibly do with your farm's records.
Start here · 13 min read
An assistant that reads your records is useful. An assistant that guesses at a withdrawal date is dangerous. The difference is architectural.
Read the guide →
Guide · 7 min read
The honest list of what a farm assistant refuses to do, written by the people who built one.
Telling a model not to look at other accounts is not a security boundary. Here is what actually is.
A guessed withdrawal date will be believed. That is what makes it worse than no answer at all.
Retrieval, not judgement — the one job an AI assistant is honestly built to do well.
A model that adds numbers will eventually add them wrong, confidently, and a wrong margin looks exactly like a right one.
What leaves your account when you ask an assistant a question, and what a scoped connection actually means.
A search box is exhaustive. A conversation is selective. Knowing which one you need saves the review.
Short, specific, sourced to a tool, and willing to say it doesn't know — a field guide to reading an answer.
An assistant is a window onto records that exist. A window onto nothing shows nothing.
An assistant that cannot act cannot act wrongly. That constraint is a feature, not a shortcoming.
All guides
