https://yanqian.github.io/tags/harness-engineering/Recent content in Harness-Engineering on Armstrong YanHugoenThu, 04 Jun 2026 10:46:50 +0800https://yanqian.github.io/posts/publish/i-built-a-small-harness-to-stop-ai-coding-projects-from-forgetting-state/Thu, 04 Jun 2026 10:46:50 +0800https://yanqian.github.io/posts/publish/i-built-a-small-harness-to-stop-ai-coding-projects-from-forgetting-state/<p>AI coding agents are powerful.</p> <p>But long-running AI coding projects break in a very specific way:</p> <ul> <li>the session is interrupted</li> <li>the context becomes too long</li> <li>the weekly quota runs out</li> <li>tomorrow&rsquo;s agent forgets yesterday&rsquo;s decisions</li> <li>the agent changes unrelated files</li> <li>the agent marks work done too early</li> </ul> <p>The problem is not that AI cannot write code.</p> <p>The problem is that AI coding projects often do not have durable project state.</p>https://yanqian.github.io/posts/publish/harness-engineering-is-about-limiting-ai-not-empowering-it/Sat, 23 May 2026 20:33:47 +0800https://yanqian.github.io/posts/publish/harness-engineering-is-about-limiting-ai-not-empowering-it/<p>Why the most important part of AI-native software engineering may not be generation, but constraint.</p> <p>This is Part 2 of the AI-Native Software Engineering series.</p> <p>It continues from <a href="https://yanqian.github.io/posts/publish/agentic-coding-mental-models-and-the-new-depth-of-software-engineering/" >AI-Native Software Engineering, Part 1: Mental Models in Agentic Coding</a>.</p> <p>The previous question was:</p> <div class="highlight"><pre tabindex="0" class="chroma"><code class="language-text" data-lang="text"><span class="line"><span class="cl">If understanding no longer comes mainly from writing code, </span></span><span class="line"><span class="cl">how do humans build mental models in an agentic workflow? </span></span></code></pre></div><p>The next question is:</p> <div class="highlight"><pre tabindex="0" class="chroma"><code class="language-text" data-lang="text"><span class="line"><span class="cl">If implementation is delegated, </span></span><span class="line"><span class="cl">where does correctness come from? </span></span></code></pre></div><p>The conversation around AI-assisted software development often focuses on one thing:</p>