Why one of our designers feels like they’re cheating when they use AI, the agent I set up to track how much of my writing is still me, and why I believe clinicians will need the same readout.
Lately, one of our designers has been feeling like they’re cheating when they use AI. A design manager who mentors them told me about it in our weekly review on Wednesday, and I hated hearing it, because this designer uses AI the way we’ve asked everyone to: to iterate faster and try more versions of an idea than they ever could by hand.
The design manager had already given them a good answer. AI is part of our workflow now, and what we’re accountable for is the outcome, whether the work solves the right problem for our customers, while how we get there is up to us. I agree with all of it, and I also know it won’t make the feeling go away, because the feeling comes from not knowing how much of the work is yours, and a principle can’t tell you that.
What I suggested instead is something I already do for myself. I’ve set up an agent that keeps track of how much of my writing is me and how much is AI. It’s a skill that gets updated with every change I make to a draft, from the single words I swap to the paragraphs I rewrite from scratch, and over time it builds a record of how much my voice is overriding the AI’s. It also gets better at sounding like me, since every edit teaches it something.
This newsletter is a good example, because most of what I write now starts as a draft from AI. The first draft of this issue kept saying “my designers” and “my design managers,” and I changed every one of them to “our,” since that’s how I talk about the people I work with. The skill logs a change like that, and the next draft starts a little closer to me.
That record is what I’d love this designer to have. When the feeling of cheating shows up, they could look at what they changed and see how much of the work is theirs, and if the answer turns out to be “not much,” they’d know where to put their attention next time. I asked the design manager to let them know they can just jump in.
Start thinking about a skill that gives you feedback on how much you’re using AI, because it’s the only safety net I can think of.
Back in Issue 18, I wrote that the obligation that survives the assistance is the choosing. A record like this makes your choices visible, to you first and to anyone who asks, which is a much better place to stand than a vague sense that you got away with something.
The same question is coming for the clinicians we design for. A clinician who uses AI to help write their notes is still signing their name to a record of someone’s care, and I believe they’re going to need the same kind of readout: a gut check on how much of the work AI did for them, and a clear signal when it starts to carry too much. That’s a design problem as much as a technical one, because the readout has to show up where the clinician is already working, in a way they’ll trust, without making them feel watched.
If you’ve felt what this designer feels, I’d start small. Keep a running note of what you change in every AI draft for a couple of weeks, and read it back before you decide how you feel about your work. I’d love to hear what you find, and I’m happy to walk you through how mine is set up.