The latest articles on DEV Community by Abdur Rafay (@abdur_rafay_ar). https://dev.to/abdur_rafay_ar 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%2F4059071%2F242eac07-5fc7-4ad2-81cc-6699d35f0c43.jpg https://dev.to/abdur_rafay_ar en Abdur Rafay Sun, 02 Aug 2026 12:27:16 +0000 https://dev.to/abdur_rafay_ar/how-i-built-relay-an-ast-based-latency-auditor-for-python-ai-agents-2jaj https://dev.to/abdur_rafay_ar/how-i-built-relay-an-ast-based-latency-auditor-for-python-ai-agents-2jaj <p>I kept running into the same problem building AI agents. <br> They were slow and I had no idea why.</p> <p>No obvious errors, logs looked fine, but requests were taking <br> way longer than they should. Turns out the codebase was full <br> of async anti-patterns. Missing awaits, sequential LLM calls <br> that could've been parallel, blocking I/O hiding inside async <br> functions.</p> <p>So I built Relay.</p> <p>It uses AST analysis to scan your agent codebase and find <br> exactly these spots. Not just flagging them, it suggests <br> fixes too. And since it runs as an MCP server, it works <br> directly inside Claude Code. No log files, no context switching.</p> <h2> how it works </h2> <ul> <li>scans your Python codebase statically</li> <li>finds async bottlenecks (missing awaits, unparallelized calls, etc)</li> <li>surfaces fixes inline in your editor</li> </ul> <h2> try it </h2> <p><em>pip install relay-arclat</em></p> <p>or check it out <a href="//relay.arclat.com">here</a></p> <p>built by Arclat, a studio making developer tools for AI agent teams.</p> agents mcp performance python