agentic-ai-engineering/site/.vitepress/dist/assets/blog_posts_vibe-vs-agentic.md.CS3ao_cg.js

import{c as t,Q as a,j as r,m as n}from"./chunks/framework.BPKcPtvA.js";const p=JSON.parse('{"title":"Vibe Coding vs Agentic Engineering","description":"","frontmatter":{},"headers":[],"relativePath":"blog/posts/vibe-vs-agentic.md","filePath":"blog/posts/vibe-vs-agentic.md","lastUpdated":null}'),i={name:"blog/posts/vibe-vs-agentic.md"};function o(s,e,d,h,c,g){return a(),r("div",null,[...e[0]||(e[0]=[n('<h1 id="vibe-coding-vs-agentic-engineering" tabindex="-1">Vibe Coding vs Agentic Engineering <a class="header-anchor" href="#vibe-coding-vs-agentic-engineering" aria-label="Permalink to &quot;Vibe Coding vs Agentic Engineering&quot;">​</a></h1><p><strong>May 26, 2026</strong></p><p>Vibe coding is prompting and shipping whatever comes out. Agentic engineering is building systems that produce reliable, verifiable results. They are not the same thing.</p><h2 id="the-5-hard-rules" tabindex="-1">The 5 Hard Rules <a class="header-anchor" href="#the-5-hard-rules" aria-label="Permalink to &quot;The 5 Hard Rules&quot;">​</a></h2><p><strong>1. Risk compounds with runtime.</strong> A 1% per-turn failure rate equals a 63% chance of disaster over 100 turns. Long-running agents are not safer. They are more exposed.</p><p><strong>2. If your agent can write AND execute code, you are back at L1 security.</strong> The marquee break: agent writes cleanup.py, runs python cleanup.py, your production data is gone. The security hook never fired because it only saw &quot;python cleanup.py.&quot;</p><p><strong>3. Every turn is a roll of the dice.</strong> Modern models refuse well 99% of the time. That last 1% grows with every feature release. Engineer the harness for the 1%, not the 99%.</p><p><strong>4. Token costs scale with loop depth, not task complexity.</strong> A simple task with a bad loop costs 10x more than a complex task with a clean loop. Optimize the loop first.</p><p><strong>5. What you cannot measure, you cannot improve.</strong> Every agent system needs: cost per task, success rate, loop efficiency, failure mode tracking.</p><h2 id="the-practical-difference" tabindex="-1">The Practical Difference <a class="header-anchor" href="#the-practical-difference" aria-label="Permalink to &quot;The Practical Difference&quot;">​</a></h2><table tabindex="0"><thead><tr><th>Dimension</th><th>Vibe Coding</th><th>Agentic Engineering</th></tr></thead><tbody><tr><td>Approach</td><td>Prompt and pray</td><td>Build and verify</td></tr><tr><td>Security</td><td>Trust the model</td><td>Trust the harness</td></tr><tr><td>Cost</td><td>Unknown until the bill arrives</td><td>Tracked and optimized</td></tr><tr><td>Quality</td><td>Whatever comes out</td><td>Measured against golden dataset</td></tr><tr><td>Iteration</td><td>Try another prompt</td><td>Fix the harness</td></tr></tbody></table><h2 id="the-karpathy-framing" tabindex="-1">The Karpathy Framing <a class="header-anchor" href="#the-karpathy-framing" aria-label="Permalink to &quot;The Karpathy Framing&quot;">​</a></h2><p>At Sequoia Ascent 2026, Andrej Karpathy said: &quot;Vibe coding raises the floor. Agentic engineering raises the ceiling.&quot; Vibe coding makes mediocre engineers productive. Agentic engineering makes great engineers extraordinary.</p><hr><p><em>From Module 1 of the <a href="/">Agentic Engineering Course</a>. The first module establishes the foundations that the rest of the course builds on.</em></p>',15)])])}const u=t(i,[["render",o]]);export{p as __pageData,u as default};