Someone asked me on a call recently whether I use AI. It was an open question, the kind that lets you answer however you like, so here’s the answer I gave him, written down where anyone can check it.
Yes. I use Claude, which is made by Anthropic. I name it because “an AI tool” is the sort of answer that makes people wonder what you’re leaving out, and if you’re deciding whether to trust someone with your data, you should know exactly what’s in the room.
AI isn’t going away, and avoiding it isn’t a sensible plan for anyone who works with data. Learning to use it reliably and responsibly is. As of this writing, I use Claude because it’s the most responsive tool I’ve tried, it’s the best at code, and I respect the ethical stands Anthropic has taken. One example: the company says it’s deeply uncertain whether Claude has anything like experiences of its own, and rather than wave that question off, it gave Claude the ability to end persistently abusive conversations, just in case. I don’t know what to make of the bigger question either. But that kind of care is what I want from the company behind a tool I trust with my work.
What it does
Website work. Every page on this site is generated by a build script, and Claude writes a lot of the basic code behind it. I read that code before it runs, I test what it produces, and nothing goes live that I haven’t checked. Code is the easy case for AI because when it’s wrong, it’s usually wrong where you can see it: the page breaks, the form doesn’t send, the layout falls apart on a phone. The mistakes announce themselves.
Design work. Layouts, color, type, three versions of a page side by side. It’s fast at giving me options to react to. Choosing between them is my job, and so is knowing when none of them is right.
Mapping and analyzing websites. Before I touch an organization’s site, I want to know every page on it, what links to what, what’s broken, and where a visitor gets lost. That’s tedious, exhaustive work, and a tool does it more thoroughly than I would at eleven at night. What to fix, and in what order, is still a conversation with the people who run the site.
My time and my to-dos. It helps me keep track of what’s due, what’s waiting on someone else, and what I said I’d do on Tuesday. That’s less glamorous than the rest of the list, and it’s probably the use that saves me the most hours.
Drafting, including this piece. Field Notes usually start with help. I rewrite, cut, correct, and argue with the draft until it says what I mean, and my name goes on it because I stand behind every sentence. If knowing that changes how you read us, that’s fair. I’d rather you know.
What it doesn’t touch
Some of the organizations I work with serve people dealing with mental health issues, including self-harm, suicide, and substance use. It shows up in their data: intake forms, check-in notes, the free-text boxes where people write things they’ve never said out loud.
Any of that data attached to a real, individual person never touches AI. Not to clean it, not to sort it, not to summarize it, and not to check my work.
That means some of my work takes longer than it would for someone with fewer rules, and some clients get less of the speed that tools make possible. I think that’s the right trade. The people in those records never agreed to have software read about their worst week, and nobody’s going to audit a small nonprofit to find out whether I respected that. So I hold the line myself.
What I mean by using it ethically
Say so. Name the tool and say what it does. If a consultant gets vague when you ask, ask again.
Check everything. Nothing a tool produces goes to a client, onto a site, or into a report until I’ve read it and would sign it. When I’m wrong, it’s my mistake, not the software’s.
Keep judgment about people human. A tool can sort a spreadsheet. It doesn’t decide who’s a good fit for a program, what a donor meant, or how someone’s doing.
Keep the most sensitive data out. See above. That one isn’t up for negotiation.
Credit people, not the tool. If an idea in a Field Note came from a person, a book, or someone else’s work, they get named. The tool doesn’t get credit for other people’s thinking.
Ask anyone, including us
If you’re hiring help with your data or your website, ask these out loud: Do you use AI? Which tool? What does it see? Who checks what it produces before it reaches us? Is there anything you won’t put into it?
A good consultant will answer all five without flinching, and a few of the answers will cost them something to say. Those are the ones to listen to.