The latest articles on DEV Community by Harry Machura (@harry_machura_67442b7a2ab). https://dev.to/harry_machura_67442b7a2ab 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%2F4060143%2F8b2fea62-1f13-4c56-a7e7-d82458bd989d.jpg https://dev.to/harry_machura_67442b7a2ab en Harry Machura Mon, 03 Aug 2026 09:48:25 +0000 https://dev.to/harry_machura_67442b7a2ab/i-built-a-language-where-ai-calls-are-sandboxed-by-default-3eg0 https://dev.to/harry_machura_67442b7a2ab/i-built-a-language-where-ai-calls-are-sandboxed-by-default-3eg0 <h1> I Built a Language Where AI Calls Are Sandboxed by Default </h1> <h2> The 30-line Python problem </h2> <p>Last month I needed a script that reads server logs, classifies errors with an LLM, summarizes them, and writes a report. In Python, it looked like this:</p> <ul> <li>Import the SDK</li> <li>Initialize the client</li> <li>Handle the API response</li> <li>Parse JSON</li> <li>Add <code>asyncio.gather()</code> because sequential calls took 8 seconds</li> <li>Write a custom sandbox because I don't trust LLMs with <code>exec</code> and file writes</li> <li>Package it in Docker because <code>requirements.txt</code> always breaks on the server</li> </ul> <p><strong>80 lines later</strong>, it worked. But it felt wrong. I wasn't building logic — I was plumbing.</p> <p>So I asked myself: <em>What if AI operations were language primitives, not library calls?</em></p> <h2> Meet Pipe </h2> <p>Pipe is a small runtime (~10 MB, single binary, zero dependencies) that treats <code>summarize</code>, <code>translate</code>, <code>classify</code>, and <code>ask</code> as first-class citizens — on the same level as <code>+</code>, <code>sort</code>, or <code>len</code>.</p> <p>Try it</p> <p>Browser Playground (WASM, no install): pipe-lang.com</p> <p>Source: github.com/MachuraHarry/pipe</p> <p>Docs: pipe-lang.com/docs</p> ai opensource programming go