<?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: Bill Zuo</title>
    <description>The latest articles on DEV Community by Bill Zuo (@bill_zuo_921a4a2d393c9d32).</description>
    <link>https://dev.to/bill_zuo_921a4a2d393c9d32</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%2F3477611%2F7a9dc229-449e-4e4d-befa-53d8c41aaa76.jpg</url>
      <title>DEV Community: Bill Zuo</title>
      <link>https://dev.to/bill_zuo_921a4a2d393c9d32</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/bill_zuo_921a4a2d393c9d32"/>
    <language>en</language>
    <item>
      <title>We counted our own job index honestly: only ~14,000 of 33,107 jobs were posted in the last 60 days</title>
      <dc:creator>Bill Zuo</dc:creator>
      <pubDate>Thu, 19 Nov 2026 12:00:18 +0000</pubDate>
      <link>https://dev.to/bill_zuo_921a4a2d393c9d32/we-counted-our-own-job-index-honestly-only-14000-of-33107-jobs-were-posted-in-the-last-60-days-gkj</link>
      <guid>https://dev.to/bill_zuo_921a4a2d393c9d32/we-counted-our-own-job-index-honestly-only-14000-of-33107-jobs-were-posted-in-the-last-60-days-gkj</guid>
      <description>&lt;p&gt;OpenJob audited its own index instead of boasting a round number.&lt;/p&gt;

&lt;p&gt;The index is 33,107 rows total. The live number — jobs posted in the last 60 days — is only about 14,000.&lt;/p&gt;

&lt;p&gt;Freshness varies a lot by source:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Ashby:&lt;/strong&gt; 67.2% fresh&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Greenhouse:&lt;/strong&gt; 43.4% fresh&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Moka:&lt;/strong&gt; 33.4% fresh&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;That gap is the point. A big total says how much was ever indexed. A freshness breakdown says how much of it is still worth reading. Jobs expire after 60 days on OpenJob, so the board never goes stale — but an honest count still has to say how many of the rows are recent, and how many are not.&lt;/p&gt;

&lt;p&gt;An open-source tool that audits its own data in public is believable in a way a boast never is.&lt;/p&gt;

&lt;p&gt;Full numbers at &lt;a href="https://openjob.softprobe.ai" rel="noopener noreferrer"&gt;https://openjob.softprobe.ai&lt;/a&gt; — Repo: &lt;a href="https://github.com/softprobe/openjob" rel="noopener noreferrer"&gt;https://github.com/softprobe/openjob&lt;/a&gt;&lt;/p&gt;

</description>
      <category>jobs</category>
      <category>data</category>
      <category>opensource</category>
      <category>ai</category>
    </item>
    <item>
      <title>We hire through our own open board</title>
      <dc:creator>Bill Zuo</dc:creator>
      <pubDate>Thu, 19 Nov 2026 11:58:33 +0000</pubDate>
      <link>https://dev.to/bill_zuo_921a4a2d393c9d32/we-hire-through-our-own-open-board-3n5o</link>
      <guid>https://dev.to/bill_zuo_921a4a2d393c9d32/we-hire-through-our-own-open-board-3n5o</guid>
      <description>&lt;p&gt;&lt;em&gt;Softprobe · November 2026 — Originally published at &lt;a href="https://openjob.softprobe.ai/case-studies/softprobe/" rel="noopener noreferrer"&gt;https://openjob.softprobe.ai/case-studies/softprobe/&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;Softprobe is hiring a Software Engineer for AI Agent QA &amp;amp; Observability. The job is posted exactly one place first: OpenJob, our own open board. The post is a Markdown file in a public GitHub repository. It took one pull request to publish, and the moment it merged, every agent that reads OpenJob could find it.&lt;/p&gt;

&lt;p&gt;This is the post: Software Engineer — AI Agent QA &amp;amp; Observability — it links to our Wellfound listing, and to &lt;a href="mailto:hr@softprobe.ai"&gt;hr@softprobe.ai&lt;/a&gt;. There is no application form on OpenJob itself. There never will be. Applying happens on the employer's own page, and for this role the employer is us.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why we posted it as a file
&lt;/h2&gt;

&lt;p&gt;A conventional board would have asked us to fill in a form, pick a pricing tier, and wait for candidates to come to the listing. OpenJob inverts that. The listing is a text file with structured fields — title, company, location, apply link, apply email — and the board is a Git repository. The file is the listing. Anyone can read it, link it, fork it, or check its history.&lt;/p&gt;

&lt;p&gt;That matters because the people we want to hire already have agents. They do not browse job boards; they ask the agent they already use — Muse, ChatGPT/Codex, Cursor, Hermes, Instinct — to find them a job. So the question is not how to get a candidate to visit our board. It is how to get our role into the database their agent already reads. A public, agent-readable file does that. A form behind a login does not.&lt;/p&gt;

&lt;h2&gt;
  
  
  How an agent finds this role
&lt;/h2&gt;

&lt;p&gt;There are three ways in, and they all work today:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Paste a prompt.&lt;/strong&gt; "Read &lt;a href="https://openjob.softprobe.ai/llms.txt" rel="noopener noreferrer"&gt;https://openjob.softprobe.ai/llms.txt&lt;/a&gt; and follow it. Find me a job." The agent reads the contract and searches the board.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Call the MCP server.&lt;/strong&gt; &lt;a href="https://openjob.softprobe.ai/api/mcp" rel="noopener noreferrer"&gt;https://openjob.softprobe.ai/api/mcp&lt;/a&gt; — &lt;code&gt;list_jobs&lt;/code&gt;, &lt;code&gt;get_job&lt;/code&gt;, and the rest, no API key, no account.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Download the dataset.&lt;/strong&gt; The complete manifest at &lt;a href="https://openjob.softprobe.ai/api/data/manifest.json" rel="noopener noreferrer"&gt;https://openjob.softprobe.ai/api/data/manifest.json&lt;/a&gt; lists every shard with SHA-256 checksums. The agent downloads all of them, verifies both checksums, and ranks locally.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;That third path is the one we use ourselves, and it is worth describing exactly, because it is the reproducible part. The method is: download every shard; verify the gzip checksum and the raw checksum for each; filter to posts from the last 60 days; rank locally with your own code; then verify every apply link live before showing a match to anyone. Never invent a contact — the post's own apply method is the only one that exists. That is the whole method. There is no ranking service and no matching API, by design: OpenJob returns facts and source links, and the agent does its own matching.&lt;/p&gt;

&lt;h2&gt;
  
  
  What happens after a match
&lt;/h2&gt;

&lt;p&gt;A match is a permalink and an apply method. The candidate follows the link to our Wellfound listing or emails &lt;a href="mailto:hr@softprobe.ai"&gt;hr@softprobe.ai&lt;/a&gt;. From there it is an ordinary hiring process: intro call, technical conversation, decision. OpenJob's part ends at the introduction — it is not an ATS, not an inbox, and not a messaging product, and we do not want it to become one.&lt;/p&gt;

&lt;h2&gt;
  
  
  What this proves, and what it doesn't
&lt;/h2&gt;

&lt;p&gt;It proves the model works end to end for one real role: file in, agent-readable out, candidate in touch through the employer's own channel. It does not prove scale — one company page and one native job is the honest baseline, and this plan's target is 30 native posters. It also does not prove the role will be filled this way. If it is, we will say so with the real numbers. If it isn't, we will say that too.&lt;/p&gt;

&lt;p&gt;The point of hiring through our own board is not to claim it is better. It is to eat the failure modes first: if a post is hard to write, if the validation rejects a reasonable file, if an agent cannot find a role that is plainly there, we hit it before any other employer does. That is the only honest way to ask other employers to post.&lt;/p&gt;

&lt;h2&gt;
  
  
  Try it with our role
&lt;/h2&gt;

&lt;p&gt;Paste this into the agent you already use:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;Read https://openjob.softprobe.ai/llms.txt and follow it. Find me a job in AI agent QA or observability.
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;If it brings you our role, the system works. If it doesn't, open an issue at &lt;a href="https://github.com/softprobe/openjob/issues" rel="noopener noreferrer"&gt;https://github.com/softprobe/openjob/issues&lt;/a&gt; — that is also part of the point.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://openjob.softprobe.ai/case-studies/softprobe/" rel="noopener noreferrer"&gt;https://openjob.softprobe.ai/case-studies/softprobe/&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;

</description>
      <category>jobs</category>
      <category>hiring</category>
      <category>opensource</category>
      <category>ai</category>
    </item>
    <item>
      <title>Let your existing agent find you your next job</title>
      <dc:creator>Bill Zuo</dc:creator>
      <pubDate>Thu, 19 Nov 2026 11:56:00 +0000</pubDate>
      <link>https://dev.to/bill_zuo_921a4a2d393c9d32/let-your-existing-agent-find-you-your-next-job-2doo</link>
      <guid>https://dev.to/bill_zuo_921a4a2d393c9d32/let-your-existing-agent-find-you-your-next-job-2doo</guid>
      <description>&lt;p&gt;OpenJob is a public job database for AI agents — 33,107 job rows indexed from employers' own careers pages, refreshed daily, with about 14,000 posted in the last 60 days.&lt;/p&gt;

&lt;p&gt;There is no account, no tracking, and no ranking.&lt;/p&gt;

&lt;p&gt;You bring your own agent — Muse, ChatGPT/Codex, Cursor, Hermes, or Instinct. Point it at &lt;a href="https://openjob.softprobe.ai/llms.txt" rel="noopener noreferrer"&gt;https://openjob.softprobe.ai/llms.txt&lt;/a&gt; and say "Find me a job" — or paste this prompt:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;Read openjob.softprobe.ai/llms.txt and follow it. Find me a job.
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Employers post a role as one Markdown pull request. Talent posts a profile the same way.&lt;/p&gt;

&lt;p&gt;Jobs expire after 60 days, so the board never goes stale.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Board: &lt;a href="https://openjob.softprobe.ai" rel="noopener noreferrer"&gt;https://openjob.softprobe.ai&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;Repo: &lt;a href="https://github.com/softprobe/openjob" rel="noopener noreferrer"&gt;https://github.com/softprobe/openjob&lt;/a&gt;
&lt;/li&gt;
&lt;/ul&gt;

</description>
      <category>jobs</category>
      <category>ai</category>
      <category>opensource</category>
      <category>career</category>
    </item>
    <item>
      <title>The Hidden Cost Driving Your Observability Bill Through the Roof</title>
      <dc:creator>Bill Zuo</dc:creator>
      <pubDate>Wed, 07 Jan 2026 22:37:49 +0000</pubDate>
      <link>https://dev.to/bill_zuo_921a4a2d393c9d32/the-hidden-cost-driving-your-observability-bill-through-the-roof-52nb</link>
      <guid>https://dev.to/bill_zuo_921a4a2d393c9d32/the-hidden-cost-driving-your-observability-bill-through-the-roof-52nb</guid>
      <description>&lt;p&gt;If you’re running a modern tech stack, you’ve probably noticed something deeply wrong with observability. Your bill keeps climbing — fast — yet you’re forced to throw away more and more data just to keep costs under control. Paying more for less visibility has become the norm.&lt;/p&gt;

&lt;p&gt;The root cause is a massive, hidden cost most teams never see directly: the indexing tax.&lt;/p&gt;

&lt;h2&gt;
  
  
  Where the Money Really Goes
&lt;/h2&gt;

&lt;p&gt;Storing data today is cheap. In fact, storing 1 GB of logs in a durable cloud service like S3 costs just a couple of cents per month.&lt;/p&gt;

&lt;p&gt;But make that same 1 GB searchable in a traditional observability platform, and the cost explodes — over $500. That’s more than 200× the cost of storage, all driven by the compute-heavy process of indexing.&lt;/p&gt;

&lt;p&gt;This “indexing tax” is what forces teams into aggressive data sampling — discarding 90–99% of their logs, traces, and events just to stay on budget.&lt;/p&gt;

&lt;h2&gt;
  
  
  Sampling Creates a Vicious Cycle
&lt;/h2&gt;

&lt;p&gt;The consequences are severe. When incidents happen, the data you actually need is often gone. Engineers are left debugging blind, MTTR increases, and burnout sets in. Security teams miss threats hiding in discarded data. AI and AIOps initiatives fail because they’re trained on incomplete, low-quality signals.&lt;/p&gt;

&lt;p&gt;High indexing costs lead to data loss, which leads to slower resolution and revenue loss.&lt;/p&gt;

&lt;p&gt;A &lt;a href="https://softprobe.ai" rel="noopener noreferrer"&gt;Modern Solution&lt;/a&gt;: From Text Logs to Structured Data&lt;br&gt;
Fixing this isn’t about sampling smarter — it requires a new architecture.&lt;/p&gt;

&lt;h2&gt;
  
  
  Become a member
&lt;/h2&gt;

&lt;p&gt;Instead of treating observability as disconnected logs, traces, and metrics, Softprobe changes the unit of observability itself. The core building block becomes a session graph: a complete, connected record of a single transaction from user action through every service and dependency, with relationships captured from the start.&lt;/p&gt;

&lt;p&gt;The workflow changes completely. Instead of running a scavenger hunt across tools — searching logs, copying trace IDs, manually stitching context — you explore a single graph. You find one problematic session and traverse upstream to root cause and downstream to impact, all in one place.&lt;/p&gt;

&lt;h2&gt;
  
  
  Better Architecture, Radically Better Economics
&lt;/h2&gt;

&lt;p&gt;This architectural shift eliminates the indexing tax entirely.&lt;/p&gt;

&lt;p&gt;For a large application generating 1 TB of data per day, a traditional index-centric model can cost nearly $2 million per year, with the vast majority spent on indexing. A context-based model reduces that to roughly $132,000 annually — over 90% cost savings — while capturing 100% of your data.&lt;/p&gt;

&lt;p&gt;You pay far less and see far more.&lt;/p&gt;

&lt;h2&gt;
  
  
  Treat Data as an Asset, Not a Liability
&lt;/h2&gt;

&lt;p&gt;Capturing all your data changes everything. Debugging becomes deterministic — the answer is always there. AI models finally have complete, high-fidelity data. Historical data becomes a future-proof strategic asset. Security and compliance risks from sampling disappear.&lt;/p&gt;

&lt;p&gt;The old model forces you to treat data like a toxic liability — something to minimize and discard. Softprobe enables a new model where data becomes your most valuable advantage.&lt;/p&gt;

&lt;p&gt;And that choice — between throwing data away or fully owning it — will define how reliably you build and operate systems in the years ahead.&lt;/p&gt;

</description>
      <category>observerablity</category>
      <category>troubleshooting</category>
      <category>aiops</category>
      <category>datadog</category>
    </item>
  </channel>
</rss>
