<?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: Samuel Osei Adjabeng</title>
    <description>The latest articles on DEV Community by Samuel Osei Adjabeng (@gigwebs).</description>
    <link>https://dev.to/gigwebs</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%2F4247273%2F1f4d8bea-a1bd-4efb-932a-72d78bfb71df.jpg</url>
      <title>DEV Community: Samuel Osei Adjabeng</title>
      <link>https://dev.to/gigwebs</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/gigwebs"/>
    <language>en</language>
    <item>
      <title>The boring-infrastructure approach to running AI agents: cron, env files, launchd, and a markdown decision log</title>
      <dc:creator>Samuel Osei Adjabeng</dc:creator>
      <pubDate>Mon, 16 Nov 2026 08:35:28 +0000</pubDate>
      <link>https://dev.to/gigwebs/the-boring-infrastructure-approach-to-running-ai-agents-cron-env-files-launchd-and-a-markdown-5738</link>
      <guid>https://dev.to/gigwebs/the-boring-infrastructure-approach-to-running-ai-agents-cron-env-files-launchd-and-a-markdown-5738</guid>
      <description>&lt;p&gt;Everyone is building agent frameworks right now. I went the other direction — the most reliable parts of my AI-agent setup are the ones with zero model involvement.&lt;/p&gt;

&lt;p&gt;I run side projects where agents (Devin, Codex, T3 Code, whatever) do most of the work. The agents were never the bottleneck. The missing pieces were operational: sessions forgot state, workers claimed things that weren't true, deterministic work burned model tokens, and human-gated steps stalled because I wasn't at the terminal.&lt;/p&gt;

&lt;p&gt;Here's the layer I actually run — all of it is boring, and that's the point.&lt;/p&gt;

&lt;h2&gt;
  
  
  Deterministic first: the model only does judgment work
&lt;/h2&gt;

&lt;p&gt;A GitHub Actions cron scans for newly funded bounty issues daily: &lt;code&gt;gh api&lt;/code&gt; search, dedup against a seen-file, commits a report, opens an issue on finds. Cost: $0/day. No model call could do this better — it's a filter, not a judgment.&lt;/p&gt;

&lt;p&gt;Rule of thumb that fell out of this: &lt;strong&gt;if the task has a spec, it's a script.&lt;/strong&gt; The model gets routed only the work that genuinely needs reasoning.&lt;/p&gt;

&lt;h2&gt;
  
  
  A state file, not a memory
&lt;/h2&gt;

&lt;p&gt;&lt;code&gt;AI_HANDOVER.md&lt;/code&gt; at repo root: verified current state only, rewritten each milestone. Every session starts by reading it, and the fix for "every session starts from zero" turned out to be a file, not a vector DB. Two rules make it work: only &lt;em&gt;verified&lt;/em&gt; state goes in (checks are cited, not asserted), and it's rewritten rather than appended, so it can't rot.&lt;/p&gt;

&lt;h2&gt;
  
  
  An append-only decision log as the audit trail
&lt;/h2&gt;

&lt;p&gt;Any irreversible action — a publish, a payout, a post — needs a logged approval in &lt;code&gt;DECISIONS.md&lt;/code&gt; first. It's a markdown file with one line per approval. When something goes sideways, the question "who approved this and why" has a grep-able answer.&lt;/p&gt;

&lt;h2&gt;
  
  
  Approvals from my phone, not my desk
&lt;/h2&gt;

&lt;p&gt;&lt;code&gt;tg-post.sh "msg"&lt;/code&gt; pushes to a Telegram bot; &lt;code&gt;tg-poll.sh&lt;/code&gt; long-polls the inbox into a log file. A launchd plist keeps the listener alive across reboots and works when the agent app is closed. ~30 lines of bash total. Slack is the same shape with a different endpoint. The effect: an agent can ask "publish this?" at 2am and I approve from my phone — the human gate without the human having to be at the keyboard.&lt;/p&gt;

&lt;h2&gt;
  
  
  Delegation contracts with verification gates
&lt;/h2&gt;

&lt;p&gt;Every delegated task gets a &lt;code&gt;WORKER_BRIEF&lt;/code&gt;: scope, files, acceptance checks. The lead agent &lt;em&gt;re-runs the checks&lt;/em&gt; before the work counts. Workers reporting "done" is a claim; the lead re-running the test is the evidence. This single habit killed most of my "looks done, isn't" failures.&lt;/p&gt;

&lt;h2&gt;
  
  
  What it is not
&lt;/h2&gt;

&lt;p&gt;It's markdown + bash + one yaml file. No framework, no SDK, no lock-in. Runner-agnostic — the same files work under Devin, Codex, Claude Code, or cron alone. Secrets live in a chmod-600 env file outside the repo; scripts take env vars, never literals.&lt;/p&gt;

&lt;p&gt;I packaged the whole layer — the doc templates, comms scripts, the bounty-radar workflow, worker-brief format — as a starter kit for people who want the files: &lt;a href="https://github.com/GiGwebs/agent-ops-starter-kit" rel="noopener noreferrer"&gt;agent-ops-starter-kit on GitHub&lt;/a&gt; (free samples in &lt;code&gt;samples/&lt;/code&gt;; full kit is a paid download, $39–149).&lt;/p&gt;

&lt;p&gt;But honestly, the two patterns worth stealing even if you never buy anything:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Verified-state handover file&lt;/strong&gt; — fixes session amnesia for $0.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Worker-brief + re-run-the-checks gate&lt;/strong&gt; — fixes worker overclaiming, same price.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Happy to go deeper on either in the comments.&lt;/p&gt;

</description>
      <category>ai</category>
      <category>devops</category>
      <category>automation</category>
      <category>beginners</category>
    </item>
  </channel>
</rss>
