Skip to content

Katon

An independent product I started, designed and shipped, with AI agents writing the code.

Role
Founder and designer
With
AI agents writing the code
At
Personal product
Status
Shipped
Year
2026

AI-written BaZi readings can get a chart's facts wrong. In Katon, my Indonesian self-discovery app, a fixed calculation decides every fact and the AI only writes the words.

My role
Founder, product owner and designer. AI agents write the code, and every product decision is mine and recorded with its date.
At a glance
  • Live at katon.app
  • Designed and directed alone, with AI agents writing the code
  • The calculation decides the facts. The AI only writes the words
  1. Ten patterns, ten cards.

    Every reading lands on one of ten patterns, and each pattern has its own card to share.

  2. The calculation owns the facts.

    Every reading comes from a fixed calculation. The AI picks the words, and a check runs before anything is shown.

    From the calculation

    Day stem
    甲 · Kayu Yang
    Pattern
    Jati · The Teak
    Traits
    TeguhMenaungiKonsisten

    The AI writes the words

    From these facts only. It cannot change a stem, a pattern or a trait.

    Checked against the facts before it is shown

  3. Two cards: a free poster, and a paid finish that shows even as a thumbnail.

    Free
    Paid
Key decisions, and why
  • Nothing locked. The free reading is complete. Paid items are extras, because asking for money before people have read anything loses them.
  • One open launch price, not a hidden price test. Friends compare screenshots, and a hidden price difference looks like cheating.
  • A plain voice. Plain Indonesian, no slang. A warmer voice was dropped after a cold read showed it made people doubt the reading was serious.