I was not losing product docs. I was losing the current decision.

오승현 — Mon, 31 Aug 2026 05:35:51 +0000

I kept blaming my notes.

A PRD would start in one place. The user flow would move to a whiteboard. Screen decisions ended up in a design file. Test criteria were buried in an AI chat. Nothing was exactly missing, but by implementation I could no longer answer a simple question: which decision is current?

That was the real reason I started building ProductChaengi.

The problem was the handoff

Most tools help create one more artifact. That part has become much easier with Codex, Claude, and ChatGPT. What did not get easier was carrying a reviewed decision into the next stage.

A feature spec might change without the user flow changing. A screen might be approved while the test case still reflected the old behavior. Then an AI coding session would start with whichever document happened to be easiest to paste.

I did not want another chat window. I wanted one project where the PRD, feature specs, personas, flows, screen designs, design system, test cases, and implementation review could stay connected. AI can draft each part, but a human decides which version becomes the current standard.

One detail I nearly turned into the whole pitch

I initially talked too much about preventing an AI from reading the same code again. It is a useful feature: a checked scope can be reused only when its file manifest still matches, and changed or directly related areas go back to review.

But that is not the whole product. The bigger job is making sure the product decision made during planning is still visible when implementation is being reviewed.

There is also a boundary I wanted to keep clear. ProductChaengi does not watch a repository by itself and does not store source contents. The connected agent calculates and sends the manifest used for comparison.

What is live, and what is not proven yet

The end-to-end artifact flow, human approval, and change-aware checkpoint behavior work in isolated verification. The public product is live.

What I do not have yet is evidence from enough outside users to say where this saves the most time. I am especially curious about solo builders and small teams that use more than one AI tool.

If that sounds like your workflow, what falls out of sync first: the PRD, the screen, the test criteria, or the implementation itself?

ProductChaengi: https://product.chaengi.com/r/dev

Disclosure: I used an AI assistant to help edit and publish this post. The product decisions and experience described here are mine, and I reviewed the final text.

DEV Community: 오승현

I was not losing product docs. I was losing the current decision.

The problem was the handoff

One detail I nearly turned into the whole pitch

What is live, and what is not proven yet