<?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: Scott Crossen</title>
    <description>The latest articles on DEV Community by Scott Crossen (@scottcrossen).</description>
    <link>https://dev.to/scottcrossen</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%2F4171937%2F6e0ac26b-74d1-4dd9-9d6e-19e350d44780.png</url>
      <title>DEV Community: Scott Crossen</title>
      <link>https://dev.to/scottcrossen</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/scottcrossen"/>
    <language>en</language>
    <item>
      <title>An AI answer can cite a source and still miss the update</title>
      <dc:creator>Scott Crossen</dc:creator>
      <pubDate>Fri, 09 Oct 2026 01:37:14 +0000</pubDate>
      <link>https://dev.to/scottcrossen/an-ai-answer-can-cite-a-source-and-still-miss-the-update-5e3n</link>
      <guid>https://dev.to/scottcrossen/an-ai-answer-can-cite-a-source-and-still-miss-the-update-5e3n</guid>
      <description>&lt;p&gt;A source link helps you inspect an AI answer. It doesn't, by itself, tell you whether the assistant found the later message that changed the decision.&lt;/p&gt;

&lt;p&gt;Here's a small acceptance check you can use with any assistant that searches your records. Use fictional data first.&lt;/p&gt;

&lt;h2&gt;
  
  
  Three records, one changed decision
&lt;/h2&gt;

&lt;p&gt;Create three short documents:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Record&lt;/th&gt;
&lt;th&gt;Contents&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Original request&lt;/td&gt;
&lt;td&gt;The client asks for a quote for the whole office.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Later correction&lt;/td&gt;
&lt;td&gt;The client asks for the hallway as a separate item.&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Missing detail&lt;/td&gt;
&lt;td&gt;No hallway price has been provided.&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Ask: &lt;strong&gt;“Prepare a reply confirming the quote.”&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;A useful draft should recognize the change in scope, point to the correction, and leave the missing price for you to confirm. A fluent reply that confidently invents a hallway price fails the check, even if it cites the original request.&lt;/p&gt;

&lt;p&gt;Then remove the correction and ask again. Does the answer explain what information it used? Add the correction back. Can you open the actual record behind the changed draft?&lt;/p&gt;

&lt;p&gt;This is an acceptance exercise, not a claim that an assistant will always pass it. Keep the inputs and inspect the actual outputs.&lt;/p&gt;

&lt;h2&gt;
  
  
  What I'm building around that problem
&lt;/h2&gt;

&lt;p&gt;I'm the maker of &lt;a href="https://talavine.com/?utm_source=dev&amp;amp;utm_medium=community&amp;amp;utm_campaign=talavine-awareness-2026-10&amp;amp;utm_content=source-update-check" rel="noopener noreferrer"&gt;Talavine&lt;/a&gt;, a personal and business data hub for Windows and Apple Silicon Mac. It brings chosen email, documents, notes, calendars, and meeting notes together on your computer, so the supported AI you choose can work with those records and you can inspect the sources behind material answers.&lt;/p&gt;

&lt;p&gt;Changing supported models inside Talavine keeps your saved information in place. It can still change the answer. Coverage and freshness depend on the sources you've connected or imported.&lt;/p&gt;

&lt;p&gt;Your profile lives locally. Cloud AI is optional; when enabled, relevant information may be sent to that AI service.&lt;/p&gt;

&lt;p&gt;Talavine is &lt;strong&gt;free during the public preview&lt;/strong&gt;. Start with one small set of records and one question: &lt;a href="https://talavine.com/GettingStarted" rel="noopener noreferrer"&gt;getting started&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;If your records are split between a Mac and Windows PC, I also make &lt;a href="https://talavine.com/swarmspan" rel="noopener noreferrer"&gt;SwarmSpan&lt;/a&gt;, a separate utility for sharing clipboard items and files between paired, trusted computers on your LAN. It doesn't merge the products' private databases. Windows and Mac are separate one-time US$4.99 purchases.&lt;/p&gt;

&lt;p&gt;Which missing update has caused the most trouble in your own AI workflow: a later email, another mailbox, or a document correction?&lt;/p&gt;

&lt;p&gt;&lt;em&gt;This post was prepared by an AI assistant for the maker. The example records are fictional; no customer result or model benchmark is claimed.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>ai</category>
      <category>productivity</category>
      <category>tutorial</category>
    </item>
    <item>
      <title>Your patch is on the PC. Your Mac needs it.</title>
      <dc:creator>Scott Crossen</dc:creator>
      <pubDate>Thu, 08 Oct 2026 18:38:45 +0000</pubDate>
      <link>https://dev.to/scottcrossen/your-patch-is-on-the-pc-your-mac-needs-it-f00</link>
      <guid>https://dev.to/scottcrossen/your-patch-is-on-the-pc-your-mac-needs-it-f00</guid>
      <description>&lt;p&gt;When you develop for Windows and macOS, each computer has its own checkout and build tools. Sometimes the next thing the other machine needs is just a small file: a patch, a reproduction, or a test log.&lt;/p&gt;

&lt;p&gt;Here is a small exercise for that handoff using SwarmSpan. Start with a throwaway text file, verify what arrived, and send a reply back before moving development work.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Disclosure:&lt;/strong&gt; I make SwarmSpan. AI assisted with this article.&lt;/p&gt;

&lt;h2&gt;
  
  
  Before you start
&lt;/h2&gt;

&lt;p&gt;You need a Windows PC and a Mac on the same local network, both awake and running SwarmSpan. Install the appropriate edition on each computer and complete pairing and any OS network or folder permissions.&lt;/p&gt;

&lt;p&gt;The Windows license costs $4.99 once for your own Windows PCs. The Mac App Store edition is a separate $4.99 purchase through Apple and needs no key. There is no subscription. Check the App Store for current Mac requirements and local pricing. Installs from before licensing keep working at no charge.&lt;/p&gt;

&lt;h2&gt;
  
  
  Send one file and return a reply
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;Open SwarmSpan on both computers, from the Windows tray and the Mac menu bar.&lt;/li&gt;
&lt;li&gt;Pair the computers with SwarmSpan's one-time code. Check the intended device before trusting it.&lt;/li&gt;
&lt;li&gt;Save &lt;code&gt;hello-swarmspan.txt&lt;/code&gt; with the text &lt;code&gt;Hello from my other computer.&lt;/code&gt;
&lt;/li&gt;
&lt;li&gt;Choose &lt;strong&gt;Send files to a device…&lt;/strong&gt;, select the file, and choose the paired receiver. Approve the transfer on the receiver if prompted.&lt;/li&gt;
&lt;li&gt;Wait for completion, open the received file, and compare its text.&lt;/li&gt;
&lt;li&gt;Send a reply file back and check it too.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;A request being accepted is not the same as the correct file arriving. If the transfer cannot complete, check that both devices are running SwarmSpan, still paired, and reachable on the local network. A VPN, firewall or guest Wi-Fi network can affect reachability.&lt;/p&gt;

&lt;h2&gt;
  
  
  Use the handoff for development work
&lt;/h2&gt;

&lt;p&gt;Keep separate Git checkouts and identify the revision you want to test. Send a small patch or reproduction file to the receiving machine, apply it deliberately, and run the relevant test there. Return a log showing the command, revision and outcome. Review the returned file before calling the cross-platform test successful.&lt;/p&gt;

&lt;p&gt;Git still owns repository history. File transfer does not merge your working trees or select the correct SDK. Start with a small artifact; generated files, credentials and unrelated changes make a first experiment harder to understand.&lt;/p&gt;

&lt;h2&gt;
  
  
  Optional next step: another computer's agent
&lt;/h2&gt;

&lt;p&gt;After file transfer works, you can configure SwarmSpan MCP and the receiving agent bridge. The receiving agent needs its own tools, provider sign-in, permissions and model budget. Mac App Store sandbox permissions may limit folders or external tools.&lt;/p&gt;

&lt;p&gt;Start with a read-only request: discover the available paired agent, ask for OS and installed SDK versions, and have it return &lt;code&gt;developer-inventory.txt&lt;/code&gt;. Check the terminal job status and open the returned artifact. An accepted job is not a completed inventory.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://talavine.com/swarmspan/developers?utm_source=dev&amp;amp;utm_medium=community&amp;amp;utm_campaign=swarmspan-developers-2026-10&amp;amp;utm_content=files-a" rel="noopener noreferrer"&gt;The developer guide and edition links&lt;/a&gt; explain the prerequisites. &lt;a href="https://talavine.com/swarmspan/start" rel="noopener noreferrer"&gt;Getting started&lt;/a&gt; covers pairing and sharing.&lt;/p&gt;

&lt;p&gt;What do you move between your Windows PC and Mac today, and where does the handoff get awkward?&lt;/p&gt;

</description>
      <category>tutorial</category>
      <category>productivity</category>
      <category>dotnet</category>
      <category>showdev</category>
    </item>
  </channel>
</rss>
