A feature that worked but didn’t feel safe, the pause a squad chose on their own, and the question I’m adding to every product review as automation grows.
Last week a feature we shipped did exactly what it was built to do, and customers still didn’t trust it. Autosave in notes was saving, and the timestamp said so. But a thread started growing in our customer community, clinicians telling each other they weren’t sure their work was safe, and the more of it I read, the clearer the gap got: “last saved” answers when the note was saved, and the question a clinician has in the middle of writing is whether their work is safe right now.
The squad that owns it paused the rollout on their own and added a saving state, the small indicator that tells you something is happening while it happens. I called it out in my “Monday Design Org Warm-Up” ritual because it’s an easy move to skip. The feature worked, the number of comments was small, and there was a perfectly defensible case for pushing through. They listened and learned instead of defending the mechanics, and the fix was small.
Every product team is adding automation right now, and we’re no different: more things that happen on a clinician’s behalf, in the background, without a button press. If that automation isn’t trusted, we haven’t finished the job, no matter what the logs say. A clinical leader on our team said something similar to the whole company this week, from the care side. The biggest risk she sees with AI in clinical work is the gap between how fast the tools get built and how slowly trust gets built. Clinicians have heard a lot of “it’s handled” from software over the years, and being careful is a reasonable response to that history.
Let’s make sure people know it works, and beyond it working, that clinicians trust that it worked.
The same idea showed up the next day in design crit, from a totally different direction. A designer brought a flow where a reviewer approves a stack of documents, one after another, and the room kept pulling apart 2 actions the first version had combined into one button: approve, and go to the next one. One of our senior designers explained why in a way I loved. Even when you sign something on paper, there’s a stamp, and the stamp is how you know you did something. Put approve and next on one button and the screen changes before the person has seen that anything landed. It’s the autosave lesson again, in a different part of the product: the action worked, and the person never got the moment that told them so.
So I’m adding a question to every product review I sit in for the rest of the year. We’ve been good at asking what a feature does for clinicians. The new one is, “Do clinicians understand and trust it?” The second half carries more weight than it looks like it should, because a person can understand exactly what a feature does and still not believe it did it this time.
If you’re designing anything that acts on someone’s behalf, I’d ask 2 questions of it before you ship. What does the person see while it’s happening? And what do they see right after, before they move on? If the answer to either one is “nothing,” the design isn’t done yet, even if the feature is.