2
1 Comment

I replaced my keyword rule engine with one AI call — and deleted half the product

Findest files your email attachments into your own cloud storage. Invoices, payslips, signed contracts — they land in the right folder without you touching them.

The original rule builder had four filter fields: sender contains, subject contains, filename contains, document type. It felt powerful. It was actually a trap — the filters matched on substrings, so a rule named "electricity bills" caught every single document labelled a bill, including a Make.com receipt that ended up filed with my power company invoices.

I deleted all four fields. A rule is now one sentence: "electricity bills for the office". An AI model reads the full attachment and decides which of your rules it belongs to — all of them evaluated together in a single call, because asking rule-by-rule makes the model agree with the first thing that looks close.

Two things I didn't expect:

1. The UI got dramatically simpler. Four inputs became one textarea, and the feature stopped needing an explanation.

2. It stopped being a paid-tier feature. When AI descriptions became the only way to write a rule, gating them behind the top plan would have left the free tier unable to create any rule at all. So the paid differentiator moved to something users actually ask for: filing the same document into several clouds at once.

Free tier, no card: findest.site. Files to Drive, OneDrive, Dropbox, Gmail and Outlook.

What's the feature you shipped that turned out to be the thing making the product worse?

posted toAvatar for product Findest
Findest
  1. 1

    The strongest part is realizing the rule builder itself was becoming the problem. Replacing several brittle filters with one natural-language rule also seems to have made the product easier to understand, not just technically simpler.