<?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: umar ahad uddin ahmed usmani </title>
    <description>The latest articles on DEV Community by umar ahad uddin ahmed usmani  (@umarahad2005).</description>
    <link>https://dev.to/umarahad2005</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%2F3920558%2F939ce413-d2a4-4c83-aa84-c0247e3b6cf7.jpeg</url>
      <title>DEV Community: umar ahad uddin ahmed usmani </title>
      <link>https://dev.to/umarahad2005</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/umarahad2005"/>
    <language>en</language>
    <item>
      <title>I let my AI agent run our Meta Ads Manager. Screenshots kept freezing — page text didn't.</title>
      <dc:creator>umar ahad uddin ahmed usmani </dc:creator>
      <pubDate>Tue, 17 Nov 2026 17:40:00 +0000</pubDate>
      <link>https://dev.to/umarahad2005/i-let-my-ai-agent-run-our-meta-ads-manager-screenshots-kept-freezing-page-text-didnt-30kc</link>
      <guid>https://dev.to/umarahad2005/i-let-my-ai-agent-run-our-meta-ads-manager-screenshots-kept-freezing-page-text-didnt-30kc</guid>
      <description>&lt;p&gt;I run ads for a small travel business. I started using Claude Code with Claude in Chrome to do real work inside &lt;strong&gt;our logged-in Meta Ads Manager&lt;/strong&gt;: read results by date, split location targeting between two ad sets, publish, and collect the ad post links for our team.&lt;/p&gt;

&lt;p&gt;The usual agent loop is &lt;em&gt;screenshot → guess → click → screenshot&lt;/em&gt;. On Ads Manager's heavy editor, the screenshot call &lt;strong&gt;timed out at 30 seconds six times&lt;/strong&gt; in one session. What kept working was reading the page as &lt;strong&gt;text&lt;/strong&gt; and the &lt;strong&gt;accessibility tree&lt;/strong&gt; — the same structure screen readers and Windows UI Automation use. The agent found the "Remove" button for one place by name, clicked it by reference, read the list back as text, and published.&lt;/p&gt;

&lt;p&gt;So I packaged that habit into &lt;strong&gt;ChromeFast&lt;/strong&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  Not a new idea — a new focus
&lt;/h2&gt;

&lt;p&gt;Reading the accessibility tree isn't new: Microsoft's &lt;strong&gt;Playwright MCP&lt;/strong&gt; is built on accessibility snapshots. ChromeFast builds on that idea for a specific job: &lt;strong&gt;your real logged-in browser and business dashboards&lt;/strong&gt;, with one rule set that works in Claude Code, Codex, Cursor and Antigravity, plus safety gates before anything that spends money or publishes.&lt;/p&gt;

&lt;h2&gt;
  
  
  The numbers (reproducible)
&lt;/h2&gt;

&lt;p&gt;I benchmarked 5 public pages in real Chrome:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Reading the &lt;strong&gt;whole page as text&lt;/strong&gt; cost &lt;strong&gt;2.6×–37× fewer tokens&lt;/strong&gt; than the screenshots needed to see it, and was &lt;strong&gt;5.6×–27× faster&lt;/strong&gt; to capture than a single screenshot.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Locating one button&lt;/strong&gt; by role/name cost &lt;strong&gt;~9–470 tokens&lt;/strong&gt;, versus ~1,366 for every screenshot.&lt;/li&gt;
&lt;li&gt;Honest surprise: dumping the &lt;strong&gt;full tree&lt;/strong&gt; can cost &lt;em&gt;more&lt;/em&gt; than a screenshot (10k–70k tokens on big pages). So the rule is &lt;strong&gt;text first, then a scoped find — never the whole tree.&lt;/strong&gt;
&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  What's in the repo
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;Claude Code &lt;strong&gt;skill + plugin&lt;/strong&gt;, &lt;code&gt;page-reader&lt;/code&gt; and &lt;code&gt;page-operator&lt;/code&gt; agents, &lt;code&gt;/chromefast:read&lt;/code&gt; and &lt;code&gt;/chromefast:do&lt;/code&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Dashboard recipes&lt;/strong&gt;: Meta Ads Manager (live-tested), Google Ads, GA4, Search Console, Shopify admin (testers wanted)&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;AGENTS.md&lt;/code&gt; for Codex/Antigravity/Windsurf, Cursor rules, installers&lt;/li&gt;
&lt;li&gt;the benchmark, so you can check the numbers yourself&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;👉 &lt;strong&gt;&lt;a href="https://github.com/umarahad2005/chromefast" rel="noopener noreferrer"&gt;https://github.com/umarahad2005/chromefast&lt;/a&gt;&lt;/strong&gt; — if it saves you time, a ⭐ helps others find it. And if you can test a 🧪 recipe on your own dashboard, I'd love a PR.&lt;/p&gt;

&lt;p&gt;&lt;em&gt;I wrote this with help from Claude (AI); the benchmark numbers come from a script anyone can rerun.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>ai</category>
      <category>claude</category>
      <category>opensource</category>
      <category>productivity</category>
    </item>
  </channel>
</rss>
