Seven days coaching a vibe coding AX team, and the hardest part was not the coding

Seven days coaching a hands-on vibe coding AX team project at a major Korean enterprise — and the hardest part wasn't the coding.

Back in university, a course called Venture Business had an unusual main assignment: go to an electronics store and interview walk-in customers for market research. It felt awkward at the time, and I couldn't see the point. This program brought that assignment back to mind.

Coding was not the hurdle. Every team built a working website within minutes through Claude — no VS Code, not a single line of source code, natural language only.

What stayed unresolved until the final day was a different question: does this app actually help its users?

That's where the old assignment finally made sense. You have to start from the pain points real users describe — not from what you think is a good idea. Even better when the builder is a real user of the problem.

If I were designing a 5-day vibe coding course, I'd spend the first 3 days almost entirely on validating whether the app is worth building. Quick mockups as conversation starters, yes — but hand non-developers a vibe coding tool too early, and the energy drifts toward UI tweaks, colors, and animations. A genuinely useful internal tool can be a blank page with one button.

What matters is whether it produces an outcome worth iterating on. That means designing the invisible part first — what data to use, how it's processed, how it's delivered — and borrowing what market researchers have always done: qualitative methods (focus groups, in-depth interviews, observation) to find the WHY, and quantitative methods (surveys, A/B tests, desk research) to measure the HOW MUCH.

Boris Cherny, who leads Claude Code at Anthropic, has said coding is largely solved. It reminds me of electricity: most of us enjoy its benefits without knowing how it works, and nobody attends a 7-month bootcamp to turn on a light at home.

But one question stays unsolved, and it's ours to answer: in which room, for what purpose, and for whom are we turning on the light? Without that answer, it's a zero value-add hobby. With it, it's a value-added product.

Aim before you advance. If the destination isn't set from the start, the harder you march, the farther you drift from it. Calibrate the compass, turn on the navigation, and make sure the guidance is right — that setup has to come first.

The real challenge isn't mastering every new technical capability of LLMs — skills, design tools, and whatever comes next. What must be certain first is whether the output serves your purpose. Otherwise the app turns useless, and nobody will maintain it.

Curious how others structure the discovery phase in their AI training programs.