I recently built a small internal app with Claude Code to solve a longtime pain we’ve had. The results were great in the end, but Claude’s plan had one big flaw. The same one I see a lot of our clients running into.
Here’s what happened…
Our advanced product owner course is a 6-week program with homework between sessions to apply the content to participants’ real work. I love the learning outcomes from the homework. And ever since we started teaching this way, I’ve hated the experience of grading the homework. Not the actual grading—I enjoy seeing students applying what they’ve learned—but the user experience around grading in our LMS (learning management system). It’s painful.
For years, I’ve talked about building a little app to improve the grading workflow. And for years, I’ve put it off because it’s never bad enough to be worth the investment.
Claude Code changes this calculus. I can build a custom mini-app like this in hours rather than days or weeks.
So, I brought this problem to Claude. I described the problem and my vision for a solution. The robot did its crunching and spit out a detailed plan.
Feature #1: Build a complete wrapper for the LMS API.
Classic. I’ve watched humans do this for 20 years. “First, we’ll build out all the infrastructure and scaffolding. Then, we can build the real features on top of it.”
The plan looked completely reasonable. But it was a trap.
Wrapping the API is a useful architectural component in this app. Is it the core value or the complex part of the work? Not at all.
The big value was in a better grading workflow. And the big, complex questions were what that better workflow would look like and whether we could build it.
I did a quick round of Feature Mining to identify a real first feature. I gave my feature to Claude, and asked it to restructure the plan around incrementally producing value and incrementally growing the architecture. Which it did quite well.
A half hour later, I was using that first feature and sharing my feedback to refine the plan.
This is a pattern I’m seeing over and over again with clients and on my own projects. AI tools (like humans) naturally treat work as complicated and make big plans for it. Instead of fighting the tool or falling into the trap, shape a first slice yourself. Use Feature Mining to find a first slice that gets you early value and learning, that probes your core complexity.
That first slice is where your experience, preferences, intuition, and taste make the biggest difference. Shape it yourself. Then, as the work moves from complex to complicated, let your AI tools do what they’re good at: generating variations, proposing plans, and suggesting the features you might have missed.
Want your team to get good at finding that first slice? Feature Mining is one of the core techniques in our vertical slicing workshop, practiced hands-on with your team’s real work. If your team is building faster with AI and the work-shaping side needs to catch up, contact us to talk about whether it’s a fit.
Last updated