The first reason people left my AI regex tool: they couldn’t edit the pattern
Hey IH —
Forge Regex generates JS regex from plain English. For a while the result was a pretty box you could copy, not a tester you could poke.
That was a bounce to regex101. I finally made the pattern a real editor.
What shipped
- Edit the
/pattern/flagsstring in place. Invalid syntax fails soft; matches/replace/export update live - Three refine buttons: Tighten, Loosen, Value only — regenerates from the current pattern + original prompt
- Failed generate/refine does not burn a credit. A successful refine does (1 gen). Pro is unlimited
Why this before more marketing
IH comments on the launch post were pretty consistent: generation isn’t the dealbreaker, confidence is. If you can’t tweak the pattern on the same page, the AI output is a suggestion, not a tool.
Still missing
Must-match / must-not example lists. That’s next. One sample string still isn’t enough to trust a pattern.
Ask
For codegen tools: do you let users edit the model output immediately, or do you force “regenerate until you like it”?
Same lesson here, different domain: AI output is a suggestion until the user can touch it. When our drafts became editable before going anywhere, the "is this even mine?" objection basically vanished. Regenerate loops feel like slot machines; an editable draft feels like a text editor. And the bounce-to-regex101 signal is worth more than a survey, users voted with a tab switch, not with polite words. Trust beats capability every time.
The bounce to regex101 is a strong signal.
Did making the pattern editable improve retention, or are users still mainly regenerating until they trust the result?