Hey Builder 👋

Someone went through my whole app with their own account and resume and wrote down everything that was wrong with it.

And let me tell you… they did not hold back.

The short version:

  • One review doc became 117 numbered rows, each with a screenshot and the file to change. 6 of them were problems Claude spotted in the screenshots that the reviewer never wrote down.

  • 96 rows got fixed between 4pm Saturday and just after midnight. 71 commits, in English, German and French.

  • 3 rows Claude checked and didn't change, because the problem turned out not to be real. I found that out writing this issue 🙃

Quickly, the rest of my week:

  • Careerkit: 21,974 people have used it, up from 20,996 on Nov 5.

  • ReplyBoss: 1 commit. A scheduled post that fails now emails its owner.

  • HelloBuilder: 0 commits again. Careerkit ate the week.

If Someone Just Sent You a Doc of Everything Wrong With Your App… This Is for You

I want to talk directly to the founder who just got the feedback doc. The one with screenshots and notes like "not sure what this is".

You're not bad at product. You're just too close to it.

Quick disclaimer: I'm not a UX person. I'm a solo founder with too many Claude Code sessions open. Everything below comes from the tracker file and my git log. Nothing is tested yet. That's this week.

How a Doc Became 117 Rows

The review covered every signed-in screen, from Nov 9 to 11: "UX & copy suggestions" for each one, and 32 screenshots.

On Saturday at 14:55 my first message to Claude was:

❝

can you read this from my personal gmail if not tell me why

Then the one that set up the whole day:

❝

i want you to see all images as well in it, and then build a tracker, so we get it done one by one

Why one by one? I want to see progress. A doc of suggestions never tells you how many are left. A numbered list does.

Claude read the doc through the Google Docs connector, pulled every screenshot in order and kept them local.

Steal this: don't ask your agent to summarize the review. Ask for rows: one per point, a screenshot number, a batch, a status and the file it lives in.

Batch by Kind of Fix, Not by Screen

Claude sorted the 117 rows into five batches:

  1. Broken or wrong (30): false text, two places that disagree, anything that blocks you

  2. First run (27): tell a new user what to do next

  3. Editor wording (26): tips and labels in the editor and ATS review

  4. Tools pages (16): wording and links for signed-in visitors

  5. New features (18): bigger builds. My pick, not Claude's

At the bottom of the tracker sits a short block called "For Claude sessions".

Quick math: 96 rows between 4pm and midnight is about one row every 5 minutes.

Steal this: put the rules inside the tracker itself. A fresh session reads the file, so it reads the rules too.

So… How Do You Run Your Own?

The lazy version, this week:

  1. Get someone who didn't build your app to use it for real, with their own account, and screenshot every screen.

  2. Hand the doc to your agent: "see all images as well, then build a tracker, so we get it done one by one."

  3. Ask for a screenshot number and a file path on every row, plus a mark for problems seen in a screenshot but never written down.

  4. Batch by kind of fix: broken first, then first run, then wording, then new features (yours to pick).

  5. Put the rules at the bottom of the tracker: still true? every language? guardrail line? moved to Done?

  6. Then type "keep going".

If you take one thing away: get someone who didn't build your app to use it. Six of the 117 problems were only in the screenshots. Nobody who knows the app sees them.

🧠 What I Learned on X

📰 AI News

🛠 Tools to Try

⚡ Skills to Try

📈 Growth & Money

💬 Builder Discussions

Okay, that's it from me. Reply with your question, I answer one every week. And if someone has sent you a review doc you haven't opened yet… open it. Then tell me the row that stung 👀

Keep shipping,

Nishant