We broke production with a one-line prompt change. Here's what we built after.
Last quarter one of our engineers made a small edit to a system prompt. Pushed it directly. No review, no history, nothing.
Within an hour our AI was responding to users with completely wrong answers.
We had no idea what changed. No diff to look at. No rollback button. Just three of us staring at the codebase trying to reverse-engineer a single line edit that had already been overwritten.
Four hours later we found it.
That night I started thinking - we would never ship code this way. Every code change has a PR, a review, a history. Why are our prompts living in a comment in a config file?
That question is why PromptOT exists.
We have been building it for a few months now - typed blocks, version history, team collaboration, evaluate across models, ship via API without touching the codebase.
The boring infrastructure stuff that prompts never had but always needed.
Launching on Product Hunt next Wednesday.
Before we do - genuinely curious. How are you managing prompts right now? Hardcoded strings, Notion docs, a shared Google Doc somewhere? No judgment. We have seen everything.


Replies