<?xml version="1.0" encoding="utf-8"?>
<feed xmlns="http://www.w3.org/2005/Atom" xml:lang="en">
  
  <title>Brief — Blog</title>
  <subtitle>Notes from building Brief.</subtitle>
  <link href="https://dobrief.app/blog/feed.xml" rel="self" />
  <link href="https://dobrief.app/blog/" />
  <updated>2026-10-13T00:00:00Z</updated>
  <id>https://dobrief.app/blog/</id>
  <author>
    <name>Runbear</name>
  </author>
  <entry>
    <title>Exhausted by verbose AI answers</title>
    <link href="https://dobrief.app/blog/exhausted-by-verbose-ai-answers/" />
    <updated>2026-10-13T00:00:00Z</updated>
    <id>https://dobrief.app/blog/exhausted-by-verbose-ai-answers/</id>
    <content type="html">&lt;p&gt;My day goes like this now. I ask my AI a simple question and get back a wall of paragraphs, three headers and a summary of the summary. Then a colleague shares a long document, and by the second page it&#39;s obvious an AI wrote most of it. Every point is made three times. Every section ends by restating itself.&lt;/p&gt;
&lt;p&gt;I&#39;m tired. Not of AI. Of reading it.&lt;/p&gt;
&lt;h2&gt;Don&#39;t be a meat proxy&lt;/h2&gt;
&lt;p&gt;The easy way out is to stop reading. Skim the answer and paste it into Slack. Forward the document with &amp;quot;looks good&amp;quot;. You become a meat proxy: a person who carries AI text from one place to another without it ever passing through their head.&lt;/p&gt;
&lt;p&gt;That doesn&#39;t work. The answer from my AI is the one I&#39;m about to act on. The document from my colleague is the decision my team is about to make. I need to understand both, and I can&#39;t afford to read every word of both, every day. My time is worth more than that, and so is yours.&lt;/p&gt;
&lt;h2&gt;Why not just ask for shorter answers?&lt;/h2&gt;
&lt;p&gt;For your own chats, you can, and it helps for a turn or two. But a model writes for the next turn of the conversation, not for you. It restates the question, covers cases you didn&#39;t ask about and thinks out loud, because models do better work when they write more.&lt;/p&gt;
&lt;p&gt;And when someone pastes their AI&#39;s answer at you, there&#39;s nobody to ask. The text is already written. Whatever you do about it has to happen on your side of the screen.&lt;/p&gt;
&lt;h2&gt;So I built Brief&lt;/h2&gt;
&lt;p&gt;Brief is a Mac app with one key. Press &lt;kbd&gt;&lt;span class=&quot;ctl&quot; aria-hidden=&quot;true&quot;&gt;⌃&lt;/span&gt;&lt;span class=&quot;sr-only&quot;&gt;Control-&lt;/span&gt;X&lt;/kbd&gt; on any window, whether it&#39;s an AI chat, a shared doc or a long thread, and it gives you the core in a few lines. In a chat app it briefs the latest exchange, the question and its answer, rather than the whole conversation:&lt;/p&gt;
&lt;figure class=&quot;status&quot;&gt;&lt;span class=&quot;tl partial&quot; aria-hidden=&quot;true&quot;&gt;~&lt;/span&gt;&lt;span&gt;ChatGPT · Sonnet 5.5 · chat · latest exchange (approx) · 8504 of 14825 chars&lt;/span&gt;&lt;/figure&gt;
&lt;p&gt;That line under every brief says exactly what it read, so you know what you skipped. When something matters, the full text is still right there, and now you know where to look.&lt;/p&gt;
&lt;p&gt;The same key does two more things. Press &lt;kbd&gt;&lt;span aria-hidden=&quot;true&quot;&gt;&lt;span class=&quot;ctl&quot;&gt;⌃&lt;/span&gt;X&lt;span class=&quot;ctl&quot;&gt;⌃&lt;/span&gt;X&lt;/span&gt;&lt;span class=&quot;sr-only&quot;&gt;Control-X twice&lt;/span&gt;&lt;/kbd&gt; and what &lt;kbd&gt;&lt;span class=&quot;ctl&quot; aria-hidden=&quot;true&quot;&gt;⌃&lt;/span&gt;&lt;span class=&quot;sr-only&quot;&gt;Control-&lt;/span&gt;X&lt;/kbd&gt; read comes back in your language. Select a paragraph you&#39;re writing and press &lt;kbd&gt;&lt;span class=&quot;ctl&quot; aria-hidden=&quot;true&quot;&gt;⌃&lt;/span&gt;&lt;span class=&quot;sr-only&quot;&gt;Control-&lt;/span&gt;X&lt;/kbd&gt;, and it proofreads or rewrites it in place. ⌘Z undoes it if you change your mind.&lt;/p&gt;
&lt;h2&gt;Who it&#39;s for&lt;/h2&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;People who work in AI chats all day.&lt;/strong&gt; You asked; the AI answered at length. &lt;kbd&gt;&lt;span class=&quot;ctl&quot; aria-hidden=&quot;true&quot;&gt;⌃&lt;/span&gt;&lt;span class=&quot;sr-only&quot;&gt;Control-&lt;/span&gt;X&lt;/kbd&gt; gives you what to do next without scrolling back through it.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;People on the receiving end.&lt;/strong&gt; The AI-written spec in your inbox, the pasted answer in a Slack thread, the report that&#39;s mostly preamble. You didn&#39;t ask for the length, and you can&#39;t ask for less.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;People who sign off on things.&lt;/strong&gt; If your job is to read a proposal and say yes or no, you need its claims, numbers and asks, and to know nothing was left out of what you read.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;People reading in a second language.&lt;/strong&gt; Twelve paragraphs of fluent English cost more when English isn&#39;t your first language. A brief, translated, costs a few lines.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;It&#39;s not for text you enjoy reading. A good essay is the point of itself. Brief is for the pages you&#39;d otherwise skim and hope.&lt;/p&gt;
&lt;h2&gt;What it won&#39;t do&lt;/h2&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;It won&#39;t watch your screen.&lt;/strong&gt; It reads one window, the one in front, and only when you press &lt;kbd&gt;&lt;span class=&quot;ctl&quot; aria-hidden=&quot;true&quot;&gt;⌃&lt;/span&gt;&lt;span class=&quot;sr-only&quot;&gt;Control-&lt;/span&gt;X&lt;/kbd&gt;. It never takes screenshots.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;It won&#39;t send your text anywhere you didn&#39;t choose.&lt;/strong&gt; It goes to the engine you pick, under your own account: your Claude or ChatGPT plan, Apple Intelligence on your Mac, or no model at all. There&#39;s no Brief server in between.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;It won&#39;t hide what it skipped.&lt;/strong&gt; A brief of a selection says &amp;quot;selection&amp;quot;. A brief of part of a page says how much. When it had to cut something you didn&#39;t ask it to, the line says so with a ⚠.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;It won&#39;t think for you.&lt;/strong&gt; It shows you the lines that matter, and the decision stays yours.&lt;/li&gt;
&lt;/ul&gt;
&lt;h2&gt;Verbose is for machines&lt;/h2&gt;
&lt;p&gt;I don&#39;t think AI should write less. Length helps a model: it reasons in the open, restates its context and covers the edge cases. When another AI reads that text next, all of it is useful.&lt;/p&gt;
&lt;p&gt;A person doesn&#39;t need all of it. You need the claim, the number and the thing you&#39;re being asked to decide. Let the AI write long, and read short. You&#39;ll move faster, and you&#39;ll still know what you&#39;re agreeing to.&lt;/p&gt;
&lt;h2&gt;Try it on this page&lt;/h2&gt;
&lt;p&gt;Brief runs on macOS 26 or later. Turn on Accessibility for it once, pick the engine you already pay for, and press &lt;kbd&gt;&lt;span class=&quot;ctl&quot; aria-hidden=&quot;true&quot;&gt;⌃&lt;/span&gt;&lt;span class=&quot;sr-only&quot;&gt;Control-&lt;/span&gt;X&lt;/kbd&gt; here. Then compare your brief with the one beside this post.&lt;/p&gt;
</content>
  </entry>
</feed>