Walk through Gangnam or Hongdae and you'll see three pilates studios per block. Sometimes four. The boom is real, and almost all of them run on the same patchwork: Naver Reservation for bookings, a KakaoTalk channel for member chat, a spreadsheet for attendance, and KakaoPay or cash for payments.

It works. Sort of. More studio owners are deciding it's not good enough.

The Naver Reservation ceiling

Off-the-shelf Korean booking tools hit a wall fast. Studio owners want their own branding. They want recurring memberships, not single-class bookings. They want equipment-specific scheduling (reformer vs mat), instructor preferences, package tracking, late-cancel penalties, automatic renewal. None of that lives natively in a generic reservation tool.

Why 'just hire a dev' stopped being the answer

A custom booking app from a Korean dev shop runs 30M to 80M won. For a single-location studio doing 20-30M won in monthly revenue, that math doesn't work. It never did.

The new option is building it themselves with AI. This is where it gets interesting, and where most people get it wrong.

The constraint problem with raw AI code

Researchers have started describing something called constraint decay in LLM agents. The short version: when you stack up too many backend constraints, the model starts contradicting itself. It breaks earlier logic. It hallucinates database schemas. Payment integration, member auth, recurring schedules, cancellation policies, refund flows, KakaoTalk notifications, instructor permissions. Each layer adds pressure. Around constraint five or six, things fall apart.

That's why tools like Cursor or Bolt feel magical on a landing page and fall over the moment you try to ship operational software.

What pre-built infrastructure actually solves

This is the part most people miss. With DontCode, the backend isn't AI-generated. The database, auth, payments, notifications, all of it is pre-configured infrastructure that already works. The AI only handles the part that's actually unique to your business. The booking flow, the membership tiers, the studio-specific rules.

So a studio owner says: 'Members buy 10-class packages. They can book classes 7 days in advance. Cancel less than 12 hours before and they lose the class. Instructors set their own availability.'

That works. Because the model isn't also trying to invent a payment processor at the same time.

What this looks like in practice

A studio in Yongsan built their member app in a weekend. KakaoPay subscriptions, automated KakaoTalk reminders, instructor schedules, member dashboards. No-show rate dropped because reminders go out without anyone touching them.

No code. No developer. No 50M won invoice.

Pilates is just the example

Hair salons, dental clinics, golf academies, taekwondo dojangs. Any business running on bookings, payments, and recurring members. Generic Korean SaaS is too rigid. Custom dev is too expensive. AI from scratch hits the constraint wall.

What works is the middle path. Pre-built infrastructure for the boring parts, AI for the business logic that's actually yours.

If you run a small business in Korea and you've been duct-taping Naver and KakaoTalk together, try DontCode. You can describe your studio (or salon, or clinic) and have something running before lunch. More posts on this kind of thing on the blog.