YC Root AccessThe Q/A Layer for the AI Coding Era
Harj Taggar and Weiwei Wu on momentic positions functional testing as verification layer for AI coding.
In this episode of YC Root Access, featuring Harj Taggar and Weiwei Wu, The Q/A Layer for the AI Coding Era explores momentic positions functional testing as verification layer for AI coding Momentic provides an AI-driven functional testing platform that impersonates real users in a browser to ensure critical user flows don’t break, processing over a million test runs per day for customers like Notion and Quora.
At a glance
WHAT IT’S REALLY ABOUT
Momentic positions functional testing as verification layer for AI coding
- Momentic provides an AI-driven functional testing platform that impersonates real users in a browser to ensure critical user flows don’t break, processing over a million test runs per day for customers like Notion and Quora.
- The founders argue that AI code generation is exploding output while shifting the bottleneck to verification, making independent validation more important than ever even as code quality improves.
- Momentic integrates into the developer loop (via MCP) so coding agents like Cursor/Claude Code can write and run Momentic tests during development, not just in CI before merge.
- They promote “truth-driven/spec-driven development,” where plain-English specs and success criteria become the durable source of truth and tests continuously maintain that truth as products evolve.
- The discussion also covers company-building lessons—raising a $50M Series A to scale go-to-market and engineering, hiring for adaptability and product intuition, and building a culture of radical candor on a 13-person team.
IDEAS WORTH REMEMBERING
5 ideasAI coding shifts the bottleneck from writing code to verifying behavior.
As code output scales (Cursor/Claude Code/Codex), teams still need a reliable way to confirm changes work in production-like conditions; Momentic positions itself as the high-leverage constraint remover.
Independent verification matters because agents can be confidently wrong.
The founders emphasize you can’t rely on the same coding agent to self-certify correctness; a third-party “source of truth” is needed to protect SLAs and prevent regressions reaching customers.
Functional tests should mirror user journeys, not just code structure.
Momentic focuses on end-user flows (clicking through the app, validating outcomes) to catch issues linters and code review can’t, especially for rich UIs like editors, drag-and-drop, and canvases.
Developer-loop testing becomes a first-class tool call for agents.
Via MCP, coding agents can invoke Momentic to run real browser verification while building features, tightening feedback cycles beyond traditional pre-merge QA.
Maintainability is the core differentiation versus generating Playwright/Selenium scripts.
They argue raw generated test code becomes an unmanageable pile (tens of thousands of lines) that breaks with UI changes, whereas Momentic encapsulates and auto-maintains the “source of truth” over time.
WORDS WORTH SAVING
5 quotesI would be disappointed in, you know, three to six months I’m still reviewing TypeScript or React code.
— Jeff An
I can’t really trust Claude Code or Cursor to tell me themselves. You know, I need a third… external source of truth for verifying that.
— Weiwei Wu
Momentic tests must pass before one of Notion’s engineers can merge their PR.
— Weiwei Wu
Your code is just an implementation of that source of truth.
— Weiwei Wu
We will win and destroy them all. Like, it… is inevitable.
— Weiwei Wu
QUESTIONS ANSWERED IN THIS EPISODE
5 questionsMomentic says it “impersonates a user” for functional testing—how exactly are user flows represented (natural language, recordings, structured specs), and what makes them robust to UI changes?
Momentic provides an AI-driven functional testing platform that impersonates real users in a browser to ensure critical user flows don’t break, processing over a million test runs per day for customers like Notion and Quora.
You cite <300ms average step time—what does a “step” mean operationally, and what infrastructure/architecture enables that speed at half a million runs/day for Notion?
The founders argue that AI code generation is exploding output while shifting the bottleneck to verification, making independent validation more important than ever even as code quality improves.
Where do you draw the boundary between what linters/code review should catch vs what Momentic should catch, and how do you prevent duplicate work in a modern AI-heavy CI pipeline?
Momentic integrates into the developer loop (via MCP) so coding agents like Cursor/Claude Code can write and run Momentic tests during development, not just in CI before merge.
On the “source of truth over time” claim: what mechanisms auto-update tests safely, and how do you avoid cementing unintended behavior (i.e., turning a bug into ‘truth’)?
They promote “truth-driven/spec-driven development,” where plain-English specs and success criteria become the durable source of truth and tests continuously maintain that truth as products evolve.
Notion migrated from Selenium (flaky selectors/XPath). What were the hardest workflows to port, and what failure modes still remain for Momentic on highly dynamic apps?
The discussion also covers company-building lessons—raising a $50M Series A to scale go-to-market and engineering, hiring for adaptability and product intuition, and building a culture of radical candor on a 13-person team.
Chapter Breakdown
Momentic in one sentence: the verification layer for software
Harj introduces Weiwei Wu and Jeff An, co-founders of Momentic, and they define the company’s core mission. They frame Momentic as a “verification layer” that helps ensure software works as intended at scale, already powering large product teams.
Why raise a $50M Series A now—and why Standard Capital
Weiwei explains the timing of the Series A as driven by achieving a repeatable sales motion and the need to scale engineering and go-to-market. He also describes the fundraising process and the specific appeal of Standard Capital’s peer-group model.
Testing 101 for non-engineers: why it exists and why it’s painful
The conversation grounds what “testing” means: ensuring code changes don’t break an increasingly complex application. Jeff shares firsthand experience at Robinhood trying (and failing) to enforce high test coverage and pass rates, highlighting why engineers resist testing work.
Code generation accelerates shipping—verification becomes the bottleneck
Harj points out that AI coding tools are increasing the volume of code shipped daily. Weiwei argues this creates a new bottleneck: proving the code works in production beyond linting and review.
Where linters and code review stop, and functional testing begins
Weiwei explains linters (style/pattern checks) and code review (human or AI) as upstream checks. Momentic is positioned as the downstream, user-perspective validation layer that confirms real flows work, avoiding reliance on slow pre-release “bug bashes.”
How Momentic fits into the dev stack: functional tests and agent tool-calls
Momentic runs functional tests by impersonating users and exercising real flows in the product. Jeff describes integrations (e.g., MCP) where coding agents can call Momentic during development to write/run tests and verify changes via a real browser session.
Why generic browser agents fall short: speed, complexity, and debuggability
Jeff contrasts Momentic with general-purpose browser agents: they’re slow, not optimized for testing, and hard to debug. Momentic optimizes interaction speed, supports complex UIs (rich text, drag/drop, canvases), and provides better failure diagnosis.
The future dev stack: less code review, more specs + external truth
Jeff predicts code review of implementation details will matter less as models improve, with code becoming a commodity. Engineering shifts toward writing requirements/specs and validating outcomes—creating demand for an independent verification source like Momentic.
Truth-driven (spec-driven) development: specs as the real source of truth
Weiwei lays out two philosophies: code-as-truth vs spec/truth-driven development. He argues production code can’t be the source of truth because it contains bugs; instead, detailed specs (flows, success criteria, edge cases) should be the truth, with Momentic enforcing them.
Why Momentic must be a standalone system: maintenance and evolving truth
Weiwei and Jeff argue verification can’t live solely inside coding agents. Momentic provides an independent source of truth and a system that maintains tests over time, avoiding massive brittle test suites and continuously adapting as the product changes.
Notion case study: from Selenium + manual testing to massive automated coverage
Weiwei tells the origin story: a Notion engineer tweets a desire for plain-English testing, Momentic gets recommended, and Weiwei onboards them the same night. Notion transitioned from manual testing and flaky Selenium suites to Momentic tests that gate merges at high volume.
Measuring ROI: dev hours saved vs preventing regressions and SEVs
Weiwei describes how customers quantify value, from direct engineering time savings to the more meaningful north star: incidents prevented from reaching users. The emphasis is on reliability outcomes, not just faster test authoring.
Roadmap, hiring, and culture: expanding platforms and staying adaptable
They outline product expansion (mobile/desktop support) and a focus on developer experience and deep workflow integration. They discuss hiring for adaptable, high-ownership engineers with product intuition, and a culture centered on radical candor and strong team processes.
Founder origins, teaming up, YC journey, and what keeps them driven
Weiwei and Jeff share how they moved from other career paths into engineering, then met through a mutual connection and decided to merge efforts. They recount applying to YC with only a prototype and early pilots, discuss early challenges (talent and fast-changing AI landscape), and close with ambition about Momentic’s impact and competitive drive.
EVERY SPOKEN WORD
Install uListen for AI-powered chat & search across the full episode — Get Full Transcript
Get more out of YouTube videos.
High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.
Add to Chrome