The latest articles on DEV Community by BlockAIx (@blockaix). https://dev.to/blockaix 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%2F3787462%2F6c42ea95-9cb3-4aba-a022-4174070b4ae0.png https://dev.to/blockaix en BlockAIx Mon, 23 Feb 2026 20:20:14 +0000 https://dev.to/blockaix/i-built-an-autonomous-development-runner-that-executes-a-roadmapjson-2n5k https://dev.to/blockaix/i-built-an-autonomous-development-runner-that-executes-a-roadmapjson-2n5k <p><a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2F8glpshvnkid25srlulxd.png" class="article-body-image-wrapper"><img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2F8glpshvnkid25srlulxd.png" alt="Agentik" width="631" height="221"></a></p> <p>For the past few months, I’ve been testing one idea:</p> <blockquote> <p>What if software projects were executed from a structured roadmap — automatically?</p> </blockquote> <p>Not random file generation. Not chatbot-style commits.</p> <p>I wanted a controlled loop: <strong>build → test → fix → static checks → docs</strong>, driven by one source of truth.</p> <p>So I built <strong>Agentik</strong>.</p> <p>Agentik is an open-source autonomous development runner that executes every task in <code>ROADMAP.json</code> through a dependency-aware pipeline.</p> <h2> Why this exists </h2> <p>Most AI coding tools generate code, but don’t manage execution.<br> CI manages execution, but doesn’t generate code.</p> <p>Agentik sits in the middle:</p> <ul> <li>Reads a task graph</li> <li>Runs tasks in dependency order</li> <li>Tests and auto-fixes failures</li> <li>Tracks token/cost usage</li> <li>Supports parallel-safe execution</li> <li>Resumes after interruption</li> <li>Can manage git history when git automation is enabled</li> </ul> <h2> What it runs </h2> <p>Define tasks in <code>ROADMAP.json</code> (or do it with a model using the AGENTS.md), then run:<br> </p> <div class="highlight js-code-highlight"> <pre class="highlight shell"><code>python agentik.py </code></pre> </div> <p>Pipeline:</p> <ul> <li> <strong>Build</strong> — implement code + tests</li> <li> <strong>Deps</strong> — install declared dependencies</li> <li> <strong>Test</strong> — run ecosystem tests</li> <li> <strong>Fix</strong> — patch failing tests (bounded retries)</li> <li> <strong>Static</strong> — lint/type checks</li> <li> <strong>Static Fix</strong> — patch static failures (bounded retries)</li> <li> <strong>Doc</strong> — update README</li> <li> <strong>Commit</strong> — when git management is enabled</li> <li> <strong>Deploy</strong> — optional hook if configured</li> </ul> <h2> The key insight </h2> <p>AI performs far better with structured work:</p> <ul> <li>Atomic tasks</li> <li>Explicit dependencies</li> <li>Output-scoped files</li> <li>Clear acceptance criteria</li> </ul> <p>Instead of “build a SaaS app,” you define deterministic units and execute them reliably.</p> <h2> Built-in guardrails </h2> <ul> <li>Single root task + dependency validation</li> <li>Parallel safety via disjoint outputs</li> <li>Saved state after phases (resume on <code>Ctrl-C</code>)</li> <li>Budget limits + estimated USD tracking</li> </ul> <p>Autonomy without rails is chaos. Agentik is about <strong>controllable autonomy</strong>.</p> <h2> Supported ecosystems </h2> <ul> <li>Python (<code>pytest</code> + <code>ruff</code>)</li> <li>Node/TypeScript (<code>vitest</code>/<code>jest</code> + <code>tsc</code>)</li> <li>Deno</li> <li>Go</li> <li>Rust</li> </ul> <p>You can also define your own ecosystem manually.</p> <h2> Try it </h2> <ol> <li>Install <code>opencode</code> (and Python :))</li> <li>Clone the repo</li> <li>Create <code>projects/&lt;name&gt;/ROADMAP.json</code> </li> <li>Run <code>python agentik.py</code> </li> </ol> <p>GitHub: <a href="https://github.com/BlockAIx/agentik" rel="noopener noreferrer">https://github.com/BlockAIx/agentik</a></p> <p>If you try it, I’d love feedback on failure handling, scheduling, and real-world task design patterns.</p> <p>Author: Thomas Heymelot (BlockAIx)<br> <a href="//www.linkedin.com/in/thomas-heymelot-a97a1479">LinkedIn</a></p> agents ai automation showdev