From an off-the-shelf shop to an order system we run from our phones
How a two-person florist in Kuala Lumpur went from Payload to Charm to Flair, and what AI changed along the way.
How we got here
- Part 01 · SoonOur first online shop was heavier to run than the flower business
- Part 02 · SoonTen backends before Flair v2, and only three ever served a customer
- Part 03 · SoonOn the edge and still slow: a new Postgres connection per request
- Part 04 · SoonWhy we deleted stock tracking from a flower shop's system
The 9-day rebuild
- Part 05 · SoonOne month after launch, I threw our order system away
- Part 06 · SoonBefore AI wrote a line of code, we wrote the spec. Then 211 decisions.
- Part 07 · Soon424 GitHub issues in 9 days
- Part 08 · SoonI stopped prompting AI and hired it a manager
- Part 09 · SoonBuilding from my phone, until the tokens ran out
- Part 10 · SoonNo Redis, no Kafka: our live updates are one Postgres trigger
- Part 11 · SoonPixel-perfect mockups, wrong in a real browser
- Part 12 · SoonThe 1am switchover: every test passed, but Claude couldn't connect
What it does now
- Part 13 · SoonMy team closes WhatsApp sales without opening our admin panel
- Part 14 · SoonWhy our AI concierge can't create orders
- Part 15 · SoonFleur Concierge with real customers