The Hardest Thing to Do with AI is Less
AI makes creation effortless. That's the problem. From unnecessary dashboards to over-engineered authentication systems, AI fills every gap with something that seems useful. Here's why I spent a week removing features from an app that could have taken two days.
You know what's harder to do with AI? Less.
And that's becoming a real problem for those of us trying to build something useful.
A while back I was building an AI readiness quiz for social media. In standard product manager practice, I asked Cursor to log user responses and progression through the funnel for later analysis. Cursor delivered the analytics, plus a complete admin dashboard with real-time visualizations. I was impressed. A dashboard hadn't even occurred to me. Much better than writing SQL queries later. But I was also unsettled. I hadn't asked for this.
As Blaise Pascal famously wrote, "If I had more time, I'd have written a shorter letter." It's always been easier to produce without focus, to let thoughts and ideas tumble out. Even pre-AI, refinement and curation were the hard part. But now that digital creation has become almost effortless, knowing what to leave out matters even more.
Last week I found myself building an app (as one does) for my daughter's school "train pool" (carpooling on the subway, for those outside NYC). We're new to the school, navigating an hour-long commute between Brooklyn and the Upper East Side. The logistics lived in a group chat with screenshots of schedules I'd accidentally scroll past, daily discussions about who's riding, and last-minute parent swaps. My anxiety about the whole transition was at an 11. This wasn't helping. Maybe software could.
When you first create something with AI, it's exciting to see all the additions it brings. Left unchecked, AI fills every gap with something that seems useful. Magic link authentication? Sure. Full roles and permissions system? Why not. Real-time sync with calendar APIs? Obviously.
But this app was replacing a 10-family group chat. Chaotic, yes. But dead simple.
The app took me just under a week. I could have been "done" in two days, but it takes longer to do less. Or more accurately, it takes longer to figure out what's actually needed. I spent most of that time removing features, not adding them.
The authentication system went through iterations. Magic links, forgotten passwords, a full roles system. But these are families who know each other. The only reason for auth was keeping minors' phone numbers off the open internet. Complicated user hierarchies and bank-level security weren't just overkill. They were obstacles.
The solution: PIN codes using the last four digits of your phone number. No passwords to forget or change. One "coordinator" role sets the schedule, bringing simplicity for everyone else. Done.
Every feature got the same treatment. Does this help a stressed parent figure out today's pickup? No? Cut it. Michelangelo talked about sculpture as freeing the figure already in the marble. Now you can conjure a sculpture instantly, but you still have to carve away everything that isn't the sculpture you actually need.
I've been seeing this pattern everywhere in my consulting work. A product manager told me their engineering team pushes back on AI-generated requirements docs. Haven't seen the docs, but I can guess the problem: too much. When AI makes it easy to generate comprehensive specifications, every feature gets fleshed out, whether or not it's core to the experience. Without sufficient context, AI fills gaps with plausible-sounding features that miss what's actually important. The appearance of completeness creates false confidence.
Same thing happens with client proposals. It's tempting to let AI craft these beautifully detailed documents. But clients don't want comprehensive. They want clarity about how you'll solve their specific problem.
I shared the app with the train pool families. They love it. It's simple enough, though it's early days to see if I can get everyone on board. Even if I have to sock-puppet it as a bridge between the app and the group chat for some families, it's still helping me.
And yes, I'm fighting the urge to add an SMS interface. It is more features but it also could allow the families involved to concern themselves with less of the distractions of a new app. My audience isn't technical. They're comfortable with text messages. We have the tools to meet them there. Plus, it would transform this from an app built with AI into an AI app, useful fodder for my consulting work.
But that's a tomorrow problem. This week, I'm practicing restraint.
Because in an age of digital abundance, less hasn't just become scarce. It's become the entire point.