The latest articles on DEV Community by Sabir Semerkant (@sabir_semerkant). https://dev.to/sabir_semerkant 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%2F3948368%2Ffd2dd450-622b-46e4-9f60-aad5516ae2e9.png https://dev.to/sabir_semerkant en Sabir Semerkant Sun, 24 May 2026 00:11:31 +0000 https://dev.to/sabir_semerkant/i-built-an-open-protocol-to-make-ai-coding-agents-follow-senior-engineering-workflows-11ep https://dev.to/sabir_semerkant/i-built-an-open-protocol-to-make-ai-coding-agents-follow-senior-engineering-workflows-11ep <p>AI coding agents are getting better fast, but I kept running into the same failure modes:</p> <ul> <li>they skip repository discovery</li> <li>they invent APIs, flags, or config keys</li> <li>they ignore existing project conventions</li> <li>they add unnecessary infrastructure</li> <li>they skip validation</li> <li>they treat unsupported checks as silently done</li> <li>they finish without a final codebase audit</li> </ul> <p>So I built <strong>The Polyglot Protocol</strong>.</p> <p>It is an open source senior-engineer workflow protocol for AI coding agents working across polyglot codebases.</p> <p>Repo:<br> <a href="https://github.com/sabir-gbs/the-polyglot-protocol" rel="noopener noreferrer">https://github.com/sabir-gbs/the-polyglot-protocol</a></p> <h2> What it includes </h2> <ul> <li>guidance for 22 languages</li> <li>language-selection rules</li> <li>pre-codegen checklists</li> <li>do-not-generate policies</li> <li>validation scripts</li> <li>adapters for Codex, Claude Code, and OpenCode</li> <li>post-codegen audit scoring</li> </ul> <h2> The core idea </h2> <p>Before an agent changes code, it should:</p> <ol> <li>inspect the repository</li> <li>preserve existing conventions</li> <li>choose the right language and tooling</li> <li>avoid invented APIs or dependencies</li> <li>validate the result</li> <li>document unsupported checks as explicit <code>N/A</code> </li> <li>finish with a final audit</li> </ol> <h2> Model-agnostic by design </h2> <p>I have tested the protocol with frontier models as well as models from Qwen, Grok, Kimi, MiniMax, and others.</p> <p>The point is not to depend on one model. Strong models still benefit from stricter engineering discipline, and smaller/local models benefit from clearer constraints.</p> <h2> Why open source it? </h2> <p>I want a practical, inspectable protocol that developers can adapt, critique, and improve.</p> <p>If you use Codex, Claude Code, OpenCode, Cursor, Aider, or similar agents, I’d be interested in feedback:</p> <ul> <li>Is the protocol too strict?</li> <li>What guardrails are missing?</li> <li>What should be optional vs required?</li> <li>What would make this easier to adopt?</li> </ul> ai opensource programming devtools