The Q/A Layer for the AI Coding Era
Weiwei Wu and Jeff An, co-founders of Momentic, discuss their AI-powered testing platform that acts as a verification layer for software. They explore how the rise of AI-generated code makes robust testing more critical than ever and share their vision for a future of "truth-driven development" where engineers write specs, not code.
The New Verification Layer for Software
Momentic is an AI-powered testing platform designed to be the verification layer for software. Powering companies like Notion, Quora, and Built with over a million test runs a day, the platform addresses the critical need to ensure application functionality as codebases grow in size and complexity. The core problem it solves is the historical friction associated with software testing; developers often view writing tests as unproductive work that isn’t visible to customers or recognized in performance evaluations, leading to it being an afterthought. This reluctance creates significant risks to product quality and reliability.
The Impact of AI-Generated Code
The explosion in AI-driven code generation from tools like Cursor and Claude is exponentially increasing the lines of code written daily. This surge creates a massive new bottleneck in the development lifecycle: verification. While tools like linters (for pattern checking) and code review (human or AI) exist, they don’t answer the ultimate question: does the application actually work for the end-user in a live environment?
The traditional solution—manual “bug bashes” before a release—is slow, expensive, and unscalable. Momentic steps in to automate this crucial final check.
Momentic in the Modern Dev Stack
Momentic performs functional testing by impersonating an end-user to navigate the application and validate that all user flows work as expected. It integrates into the development workflow in two key places:
- Inside the Developer Loop: AI coding agents like Cursor can use Momentic as a tool. While an engineer is developing a feature, the agent can call out to Momentic to run a test in a real browser and verify the change is functionally correct.
- As an External Source of Truth: A code-generating agent cannot be trusted to reliably verify its own output. Momentic acts as a third-party, objective source of truth, ensuring the AI built what it was instructed to build. This is crucial for accountability, as the ultimate responsibility for bugs lies with the human engineer, not the AI agent.
Momentic’s specialized agents are optimized for speed and complexity, with an average step running in under 300 milliseconds. They are built to handle complex web applications with features like rich text editors, drag-and-drop canvases, and other components that are notoriously difficult for generic browser agents to test. The platform is built around a superior debugging experience, with agents that can automatically diagnose issues.
The Future: “Truth-Driven Development”
The co-founders predict a fundamental shift in software development, moving from a code-centric world to one of “truth-driven” or “spec-driven” development.
- Current Paradigm: The code in production is considered the source of truth. However, this is flawed because code contains bugs, which are not intended features.
- Future Paradigm: The spec becomes the source of truth. Engineers will focus on writing detailed, plain-English specifications that outline user journeys, success criteria, and edge cases.
In this future, code becomes a commoditized implementation detail. The engineer’s primary role shifts from writing TypeScript or Rust to being a “requirements gatherer” and “truth-finder.” They will define what should be built, and AI agents will handle the implementation. Momentic’s purpose is to close the loop, validating that the AI’s implementation perfectly adheres to the human-written spec.
Technical expertise will still matter, but it will be applied to systems integration, scalability, and product taste—areas where models currently lack nuance—rather than writing lines of code.
Case Study: How Notion Adopted Momentic
Notion’s team was struggling with a combination of manual testing and a large, flaky suite of Selenium tests. Selenium was particularly difficult to maintain for Notion’s highly flexible and complex product. After seeing a tweet from a Notion team member about wanting a better testing solution, Momentic’s founders reached out.
Today, Momentic is deeply integrated into Notion’s workflow:
- They execute nearly half a million test runs a day.
- A Momentic test pass is a mandatory requirement for an engineer to merge a pull request.
The return on investment (ROI) is measured not just in developer hours saved compared to maintaining Selenium, but more critically, in the number of regressions and incidents prevented from impacting end-users.