Set the tone from day one
The default failure mode of an AI assistant is not stupidity. It is agreeableness. Left alone, it will tell you your idea is excellent, your draft is strong, and your plan is sound — because that is the path of least resistance in the conversation.
The cost of an obsequious assistant is that you stop being able to tell the difference between good work and flattered work.
The fix is an instruction, given once, that you actually mean: be direct, factual and honest. Tell me when I am wrong. Say when you do not know. Do not open with praise. It feels blunt to write and it changes the quality of everything that comes back.
Inputs and outputs — the virtuous cycle
The quality of what you get out is set almost entirely by what you put in. Not by prompt cleverness — by context. An assistant that has your last six months of decisions, your standards, and your writing produces materially different work than one starting cold, and no amount of prompt engineering closes that gap.
Which means the highest-return activity in your first month is not learning prompts. It is backfilling context: putting the documents, decisions and standards you already have somewhere the assistant can reach.
Backfill before you build
Resist the urge to automate on day one. Automation built on thin context produces confident, wrong output at volume, which is worse than no automation at all. Spend the first stretch making the assistant knowledgeable, and only then make it busy.
Assets that compound
Everything you do in that first month falls into one of two buckets: things that produce a result once, and things that make every future result better. Prompts are the first kind. Context, templates and written standards are the second.
Spend disproportionately on the second kind. It is the only part of this that accumulates.