Skip to main content

Use caseWrite

Revise in the margin, one comment at a time

A comment on the sentence that is wrong — not a paragraph pasted into a chat box — and the change comes back on that sentence, with the thread still open for the next round.

You start with
A draft in the folder MindMax has open, and the AI you write with: MindMax’s own, or Claude Code or Codex connected to the same folder.
Time
About forty minutes for a chapter — a few minutes marking, the rest reading what came back.
Your part
Knowing what is wrong, saying it in one line, and deciding which of the changes you actually meant.

What you'll end up with

  • A draft where only the places you marked have changed — not a rewritten document to diff by eye.
  • Every comment answered on its own passage: what it changed, or why it left that one alone.
  • A second round that starts from the threads still open, instead of from retyping the same request.
  • Comments that MindMax’s AI and a connected agent both read, so changing tools does not lose the thread.

The chat box is where you describe work. The margin is where you point at it.

From “this sentence is wrong” to a draft you accept

  1. Open the draft where you already write it

    You

    In MindMax, or in Claude Code or Codex with the same folder connected — Settings → External AI, one click. Both sides read the same comments, so it does not matter which one you start in.

  2. Mark the sentence, not the section

    You

    Select the words that are wrong and comment on them. One line is enough: “this repeats the paragraph above”, “this needs the 2024 figure, not 2022”. The comment is anchored to those words, so nobody has to work out which paragraph you meant.

  3. Pick the kind, when the words are few

    You

    A comment can be set to rewrite, to check, or to find a source. The intent travels with the comment, so “?” on a doubtful number is already a complete request — and picking one with no text at all still counts.

  4. Hand the pass over

    You ask · MindMax answers

    AskGo through the comments on the chapter I have open in MindMax and fix what they point at. Leave anything you are not sure about as a comment instead.

    You can also press the button in the comments panel that copies that sentence for you. Typing it yourself behaves the same way: the range comes from what you said, not from the button.

  5. It answers in the thread, on the sentence

    The change arrives as tracked changes on that passage, and the comment gets a reply saying what it did and why. A comment it cannot act on stays open with the reason, so you can sharpen it and send it back rather than wonder what happened.

  6. Accept the ones you meant

    You

    Decisions are per group, not per character: accept, reject, or accept and then edit by hand. Accepting marks the comments behind it done; rejecting leaves the thread open, so nothing closes quietly.

  7. The next round starts in the margin

    You ask · MindMax answers

    AskWhat is still open? Read my replies and continue from them.

    Answer one of its replies and the turn comes back to the AI — it reads the thread, not the chat history. That is how a review that would have been four meetings becomes four comments.

Why it works here

The instruction sits on the text it is about

A chat message has to describe a location: “the third paragraph”, “the bit about the trial”. A comment is already attached to the words, so the request cannot land in the wrong place — and it is still there when you open the file next month.

How a comment becomes a request →

Every round has an owner

A thread is either waiting on the AI or waiting on you, and the panel shows which. That is what stops two assistants from answering each other in circles, and what makes “what is still open” a question with an answer.

What the connection exposes →

The same comments, whichever AI you use

Comments live in the workspace, not in one app’s history. MindMax’s AI and a connected agent read the same threads, so a pass in Claude Code and a pass in MindMax continue the same conversation instead of starting a new one.

Comments and highlights →

Also used here: Guide: working in the margin with an agent · Why the instruction goes on the sentence, not in the chat box · Reviewing and revising with AI · Every tool an app can call

Where it gets hard

  • A comment points; it does not solve

    It edits the passage the comment sits on, not the document. “Make this better” gets a guess and a confident one; naming the problem — repetition, a stale number, a claim with no source — gets a fix you can judge in five seconds.

  • Read-only material takes comments, not edits

    On a PDF, an ebook or slides a comment can be answered or checked and the answer lands in the thread, but the text does not change. A PDF gets an editable Markdown copy beside it; a Word file is revised from a request in the assistant instead.

  • Someone still has to decide

    Every change waits for your accept. That is the point of the arrangement, and it is also the cost: forty comments is forty decisions, and a pass you skim is a pass you did not really make.

  • The thread is only as good as its anchor

    If you rewrite the marked sentence yourself after commenting, the comment can lose its place. It still shows the text it quoted, but the edit may land on what is there now rather than what you meant.

  • One turn at a time

    A thread is never worked by two assistants at once: whoever replies last hands the turn back to you. You can call a different app next round — it reads the same thread — but not both in the same breath.