Momentic (YC W24) Launches AI Agents to Catch Bugs Automatically

Wei-Wei and Jeff, alumni of YC W24, have launched Momentic, a platform featuring autonomous QA agents designed to plan, explore, and catch bugs across web, iOS, Android, and upcoming desktop applications. Having worked at companies like Robinhood and Retool, the founders built this tool to eliminate the pain of traditional testing frameworks like Playwright and Selenium.

• Momentic builds custom QA agents powered by the Mo framework to catch real-world bugs autonomously.
• Benchmarked against Codex, Claude Code, and Devin across four popular open-source apps.
• The Mo agent successfully caught 269 bugs during a 60-minute benchmark run.
• The company offers $250 in free spend for new signups and recently added PR testing for Vercel and Cloudflare.

The Limitations of Traditional Testing and Coding Agents

Modern software suffers from numerous bugs, and change failure rates are accelerating alongside the adoption of coding agents. While LLMs excel at generating code, they often reinforce their own assumptions, write useless tests, and struggle with visual understanding. Traditional E2E frameworks like Playwright, Selenium, or Maestro force developers to hardcode brittle implementation details such as XPath and aria attributes.

To address this, the team built Mo, a custom agent harness based on millions of test traces from complex real-world apps. Rather than writing static scripts, Mo's agents test applications dynamically. The architecture features a self-organizing agent graph (DAG), specialized domain agents for analyzing screenshots and video, and scalable infrastructure capable of spinning up hundreds of concurrent browsers.

Benchmark Results and Comparative Analysis

The Momentic team benchmarked Mo against leading coding agents on four popular open-source apps (Gitea, Metabase, Twenty, and Saleor) containing 1,396 human-labeled bugs. During a time-boxed 60-minute run, Mo detected 269 bugs with a 19.3% recall, 84% precision, at a cost of $2.67 per bug.

By comparison, Codex with Playwright MCP found 86 bugs (6.2% recall, 86% precision, $0.80 per bug), Claude Code caught 59 bugs (4.2% recall, 80% precision, $4.09 per bug), and Devin identified 27 bugs (1.9% recall, 84% precision, $3.89 per bug). Notably, Mo finds more bugs the longer it runs, increasing its bug count by 52% when test duration was extended from 30 to 60 minutes.

What This Means for Developers and Businesses

The emergence of tools like Momentic signals a paradigm shift in software quality assurance: the future of QA lies in autonomous agents testing applications themselves rather than engineers writing manual scripts. With newly shipped PR testing support for Vercel and Cloudflare deploy previews, developers can catch critical regressions before code reaches production.

For tech companies and engineering teams, leveraging autonomous QA agents significantly reduces maintenance overhead and helps uncover legacy API issues, rendering bugs, and routing logic flaws that traditional automated suites often miss.