Claude Code made the report. I still had to copy every review comment back by hand.
i use Claude Code and Codex for HTML decks and Markdown reports. making the file wasn't the slow part.
i'd send it out, get a screenshot + “the chart near the bottom”, figure out which version they meant, then paste the note back into the agent. v2 usually lived somewhere else. pretty dumb loop.
i built PreApp to get rid of that handoff. the agent publishes the HTML/MD with its local images. the reviewer gets one link to read and another to leave feedback. later the agent pulls the notes back with the exact version and location, then i choose what it should change.
i ran one HTML case from v1 to v4. 6 real notes, 2 rounds of changes, same link the whole time. there's a 41 sec recording in the repo.
i'm trying to run 3 more real, non-sensitive reports/decks this week. no polished demo needed. comment with Claude Code + HTML or Codex + Markdown and i'll help get the first link open.
The interesting opportunity isn't collecting review comments—it's preserving context so AI agents can iterate without humans rebuilding it every round. I'd keep validating whether teams adopt PreApp because it simplifies reviews or because it becomes the missing collaboration layer between AI-generated work and human decision-making.
yeah, i'm not sure yet either. when someone says “that chart”, getting the exact version + element back into the agent feels useful, but that's still a test, not adoption.
do you have an html report/deck that you send to someone else for review? i'd rather run one of yours and see which part still feels annoying after round one.
Interesting question — I don't have a specific example I can share here, but I appreciate you asking.
Would be good to continue the conversation outside the thread. What's the best email to reach you on?
yep, @serrendypity on X is easiest. just send which agent you used + what it outputs (single html, a folder, or md). no need to send the file yet. if it fits, i'll send the 3-step publish path and stay with it til the first review link opens.
I appreciate that.
I don't have a specific example I can share here, but the reason I raised the question was less about testing the workflow myself and more about the strategic decision around what PreApp becomes.
The interesting question is whether the product is mainly reducing review friction or becoming the layer that preserves context between human judgment and AI output. That distinction will likely shape who values it most.
Happy to discuss it further if useful.