<?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>Forem: Sergei Peleskov</title>
    <description>The latest articles on Forem by Sergei Peleskov (@greza_dev).</description>
    <link>https://forem.com/greza_dev</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.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F3879098%2F0b4cae54-3dc5-4228-82a2-965310ea816f.png</url>
      <title>Forem: Sergei Peleskov</title>
      <link>https://forem.com/greza_dev</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://forem.com/feed/greza_dev"/>
    <language>en</language>
    <item>
      <title>Stop Vibe Coding. Use Spec-Driven Development Instead.</title>
      <dc:creator>Sergei Peleskov</dc:creator>
      <pubDate>Tue, 14 Apr 2026 17:47:22 +0000</pubDate>
      <link>https://forem.com/greza_dev/stop-vibe-coding-use-spec-driven-development-instead-44cc</link>
      <guid>https://forem.com/greza_dev/stop-vibe-coding-use-spec-driven-development-instead-44cc</guid>
      <description>&lt;p&gt;Vibe coding is how juniors ship bugs fast.&lt;/p&gt;

&lt;p&gt;You describe a feature in natural language, the AI generates code, &lt;br&gt;
you tweak until it works. Fast? Yes. Scalable? No.&lt;/p&gt;

&lt;p&gt;At scale, vibe coding gives you:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;500 lines of unreviewable code&lt;/li&gt;
&lt;li&gt;Features you didn't ask for&lt;/li&gt;
&lt;li&gt;An agent that rewrites half your codebase&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  The fix: Spec-Driven Development
&lt;/h2&gt;

&lt;p&gt;Senior engineers don't prompt AI directly. They write specs first.&lt;/p&gt;

&lt;p&gt;Three documents you need:&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;1. PRD (Product Requirements Doc)&lt;/strong&gt; — what you're building and why&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;2. Technical Design Doc&lt;/strong&gt; — architecture, data models, API contracts&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;3. AI Spec&lt;/strong&gt; — precise instructions for the agent, edge cases included&lt;/p&gt;

&lt;p&gt;The agent works from the spec. Not from vibes.&lt;/p&gt;

&lt;h2&gt;
  
  
  Example: JWT Authentication
&lt;/h2&gt;

&lt;p&gt;Instead of prompting "add JWT auth", you write a spec that defines:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Token structure and expiry&lt;/li&gt;
&lt;li&gt;Refresh logic&lt;/li&gt;
&lt;li&gt;Error handling for each case&lt;/li&gt;
&lt;li&gt;What the agent should NOT touch&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Result: reviewable, production-grade code.&lt;/p&gt;

&lt;h2&gt;
  
  
  The 5-step workflow
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;Write PRD&lt;/li&gt;
&lt;li&gt;Write Technical Design Doc&lt;/li&gt;
&lt;li&gt;Generate AI Spec from the two above&lt;/li&gt;
&lt;li&gt;Run agent against the spec&lt;/li&gt;
&lt;li&gt;Review diff — not the entire codebase&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Full breakdown with examples in the video:&lt;br&gt;
&lt;a href="https://youtu.be/5ve3_inIN-8" rel="noopener noreferrer"&gt;https://youtu.be/5ve3_inIN-8&lt;/a&gt;&lt;/p&gt;

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