Heading into Q4 I was in a really good place, but my work felt scattered and everything Claude handed me was way too wordy. So we took a big step back, rebuilt my business plan into two pages, and cut fifteen skills down to five so the plan actually runs my week.
A simple year-end mailing list took me far longer than it should have, and I know where the buttons are. October is not scary because of the letter. It is scary because of everything underneath the letter, and that is where AI can actually help.
A few months into using AI every day, the tools you built start answering with things you retired months ago. Here is how I audited my own AI setup like any other system, and why the cleanup was mostly taking things out.
Before you plug AI into anything, you need to know which connections are yours to make. The Personal Access Map shows you where you stand in about ten minutes, and the workbook helps you build from there.
The gap between that world and this one keeps shrinking. The first time I heard about generative AI, I hated it. It scared me. Most of the fundraisers I talk to feel the same thing, and nobody has given them permission to say it out loud. Everything is moving so fast it feels like we cannot possibly keep up.
You cannot run a marathon on your first day. You can put on the shoes and walk to the end of the street. That is the whole method.
If you know AI could help but have no idea where to start, you are not behind and you are not alone. It is one of the most common things a team runs into right now, and it is rarely about the tools.
There is a meeting on your calendar that you have moved twice, and a thinking partner set up correctly will not let you look away from it. The starter prompt that makes this work closes the back door: “You are not here to validate my thinking” cuts off what avoidance is looking for in the first place.
Ninety-two percent of nonprofits are using AI. Four percent have built a workflow. Eight people in the same organization can be using the same tool in eight different ways, with no shared prompt library and no internal agreement on what data goes into a model.
