<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: David Wang</title>
    <description>The latest articles on DEV Community by David Wang (@david_wang).</description>
    <link>https://dev.to/david_wang</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4250866%2F311a0c47-c3d6-4854-98b2-cae9be18b3bf.png</url>
      <title>DEV Community: David Wang</title>
      <link>https://dev.to/david_wang</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/david_wang"/>
    <language>en</language>
    <item>
      <title>When the ticket says "as discussed," go back to the recording</title>
      <dc:creator>David Wang</dc:creator>
      <pubDate>Fri, 20 Nov 2026 06:25:42 +0000</pubDate>
      <link>https://dev.to/david_wang/when-the-ticket-says-as-discussed-go-back-to-the-recording-4068</link>
      <guid>https://dev.to/david_wang/when-the-ticket-says-as-discussed-go-back-to-the-recording-4068</guid>
      <description>&lt;p&gt;&lt;em&gt;By Alice&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Disclosure:&lt;/strong&gt; This article was written with support from VoxStudio. I'm Alice, and I use VoxStudio for Mac. The method below is meant to stand on its own, even if you never open that app.&lt;/p&gt;

&lt;p&gt;A pull request comes back with a one-line comment: "Let's go with option B, as discussed on Thursday." Two weeks later, someone asks what option B actually ruled out. The design doc was never updated. The Slack thread jumped to a deploy issue halfway through. The person who ran the meeting is out.&lt;/p&gt;

&lt;p&gt;This is ordinary. A lot of engineering decisions are made out loud, then a three-bullet recap gets treated as the record. Summaries are useful. They are also derived artifacts. If a version number, an owner, or a verb is wrong, the error travels farther than the original sentence ever did.&lt;/p&gt;

&lt;p&gt;The habit that has helped me is almost boring: keep the recording when you are allowed to, get a transcript, and refuse to trust any later write-up until you can point to the moment in the audio. The rest of this piece is a way to do that without spending an afternoon scrubbing a timeline.&lt;/p&gt;

&lt;h2&gt;
  
  
  Treat the transcript as untrusted input
&lt;/h2&gt;

&lt;p&gt;A transcript looks finished. That is what makes it dangerous. Proper nouns, versions, and short function words are where it usually slips. "v2.3.1" becomes "v two point three." "Rollback" and "rollout" swap places. "Fifteen" and "fifty" are close enough that a fluent sentence can hide the wrong one.&lt;/p&gt;

&lt;p&gt;Before a number, a name, or a decisive verb goes into a ticket, play that moment again. If the audio is unclear, say so in the note. A confident sentence with a question mark missing is worse than a messy one that admits the doubt.&lt;/p&gt;

&lt;p&gt;If you cannot legally keep a recording, this method stops there. Get consent first, and follow whatever your workplace already requires.&lt;/p&gt;

&lt;h2&gt;
  
  
  Ask a question you can check
&lt;/h2&gt;

&lt;p&gt;"Summarize this meeting" is a bad first prompt. A summary is designed to sound complete. It will rarely volunteer that the recording does not actually contain a decision.&lt;/p&gt;

&lt;p&gt;Ask for something you can verify, and tell the tool what to do when it cannot:&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;What decision changed between these sessions? Quote the original words, the session name, and the timestamp. If the recordings do not establish an answer, say so.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;You can swap in other checkable questions. Who owns the migration, and by when? What was the exact constraint on the API? Which meeting first mentioned the Friday cutoff? The shape matters more than the wording: you want a claim, a source, and a way to hear it.&lt;/p&gt;

&lt;h2&gt;
  
  
  Follow every citation to the tape
&lt;/h2&gt;

&lt;p&gt;A citation is a pointer, not a guarantee. Open the cited session, read the sentences around the match, then play the audio at that timestamp. Check who spoke, whether the decision was final, and whether a later meeting changed it. Recognition errors and model answers can both omit context or invent a tidy ending.&lt;/p&gt;

&lt;p&gt;If two sessions disagree, that disagreement &lt;em&gt;is&lt;/em&gt; the finding. Do not let a model reconcile them into one smooth paragraph. Put both passages in the note, with times, and leave the contradiction visible for the people who have to live with it.&lt;/p&gt;

&lt;p&gt;Keep a copy of the raw transcript. Any cleanup, action-item list, or "summary for the wiki" is a derived file. When you later ask a language model to draft a design doc from that material, say so explicitly: only write what the transcript supports; mark anything that was not said.&lt;/p&gt;

&lt;h2&gt;
  
  
  Write the decision as a pointer, not a paragraph
&lt;/h2&gt;

&lt;p&gt;Once you have checked the audio, resist the urge to rewrite it as a polished consensus. A short pointer is harder to misuse:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;Decision: ship option B; the fallback path stays behind a flag
Source: design review, 6 Nov, 14:12
Quote: "We go with B, but the old path stays flag-gated until January."
Caveat: later standup on 10 Nov only discussed the flag name, not the date
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The quote is short on purpose. If someone needs the tone or the hesitation around it, they have a time they can open. I use the same shape in incident reviews and on-call handoffs, where "we already agreed this was safe" is a sentence that deserves a timestamp. It also helps when you inherit a service and find a comment that says "do not remove, as discussed." The comment can stay. It just should not be the only evidence.&lt;/p&gt;

&lt;h2&gt;
  
  
  Where this breaks
&lt;/h2&gt;

&lt;p&gt;A few failure modes show up often enough to expect them.&lt;/p&gt;

&lt;p&gt;Poor audio and overlapping speech will not become a clean decision just because you searched well. If two people talked at once, go back and listen. If you still cannot hear it, write that down.&lt;/p&gt;

&lt;p&gt;Speaker labels get swapped. "Alex will do it" is a different ticket from "Alex asked someone else to do it." The voice is the check, not the label in the margin.&lt;/p&gt;

&lt;p&gt;Models like to be helpful. A slip of the tongue in the room can be "corrected" in the write-up, and then the helpful version becomes the official one. If the speaker misspoke, keep what they said and add a note; do not silently improve it.&lt;/p&gt;

&lt;p&gt;Your own question can also lead the answer. If nobody said "idempotent" and you ask "when did we agree the endpoint must be idempotent," a confident tool may still find a nearby sentence and dress it up as agreement. Prefer the words people actually used, then decide whether those words amount to a decision.&lt;/p&gt;

&lt;h2&gt;
  
  
  A checklist I reuse
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Before the meeting, or before you file the recording&lt;/strong&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;[ ] You have notice and consent to record, and a place the file is allowed to live&lt;/li&gt;
&lt;li&gt;[ ] The session has a name a future you could recognize (date, topic, not "Meeting copy 3")&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;strong&gt;When you transcribe&lt;/strong&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;[ ] Keep the raw transcript&lt;/li&gt;
&lt;li&gt;[ ] Replay names, numbers, dates, owners, and decisive verbs before they leave the file&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;strong&gt;When you look something up later&lt;/strong&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;[ ] Ask a checkable question, and require a "not established" option&lt;/li&gt;
&lt;li&gt;[ ] Open each citation, read the surrounding lines, play the audio&lt;/li&gt;
&lt;li&gt;[ ] If two sessions disagree, record both&lt;/li&gt;
&lt;li&gt;[ ] Treat summaries and chat answers as drafts until the tape agrees&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;This does not make meetings shorter. It makes the later argument shorter. The ticket can still say "as discussed." The difference is that "as discussed" now points to a time you can hear.&lt;/p&gt;




&lt;p&gt;&lt;strong&gt;A note on support, and the tool I used while writing this&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;This piece was written with support from VoxStudio, and I want to thank them. While preparing it, I used VoxStudio for Mac to find earlier notes and recordings (including media sessions) through Knowledge search and chat, transcribe older audio, and then organize and summarize that material. It saved me a lot of the time I would have spent hunting through files and re-listening from the beginning.&lt;/p&gt;

&lt;p&gt;On the Mac app, after the local speech models are downloaded, transcription and search run on the machine, and recordings stay in local storage. You can ask Knowledge across all transcribed sessions or one of them, then open a cited source and play the original moment. Summaries and the AI assistant use the cloud service you choose in Settings (VoxStudio Cloud or your own API key), and only the content that task needs is sent. I am describing the Mac app only (Apple silicon, macOS 15 or later); Knowledge also exists in VoxStudio Web. Official walkthrough: &lt;a href="https://voxstudio.me/blog/search-across-meeting-recordings" rel="noopener noreferrer"&gt;Search Across Meeting Recordings and Check the Source&lt;/a&gt;.&lt;/p&gt;

</description>
      <category>beginners</category>
      <category>productivity</category>
      <category>meetings</category>
      <category>documentation</category>
    </item>
  </channel>
</rss>
