Skip to content
Y CombinatorY Combinator

Mukund & Madhav: How Solving Verification Enabled 7M Apps

Suitebench's multi-agent verification enabled long-horizon autonomy; non-technical builders now ship production software without writing a single line of code.

MukundguestJared FriedmanhostMadhav Jhaguest
Mar 15, 202639mWatch on YouTube ↗

At a glance

WHAT IT’S REALLY ABOUT

Emergent empowers non-coders to ship production software with AI agents

  1. Emergent pivoted from AI-driven software testing to general coding agents after realizing verification loops unlock long-horizon autonomous engineering.
  2. They differentiated from “prototype-first” competitors by building an end-to-end production platform (infra, deployment, testing, security, hosting) rather than front-end demos.
  3. The company found product-market fit with non-technical users—now ~80% of users—who build serious business software, internal tools, and increasingly agentic workflows.
  4. They argue second movers can win in AI by reimagining products around new model capabilities, learning from incumbents’ gaps, and pairing superior product with aggressive distribution.
  5. Emergent positions the future as “personalized software” where niche apps, solo builders, and multi-agent systems expand software creation rather than eliminate jobs.

IDEAS WORTH REMEMBERING

5 ideas

Verification is the unlock for autonomous software engineering.

They started with test automation and concluded that strong verification loops let agents run longer and more reliably, enabling broader end-to-end engineering beyond just writing code.

Winning “vibe coding” requires production, not prototypes.

Emergent’s core bet is that users ultimately need deployment, debugging, back-end support, security, and hosting; optimizing only for quick UI prototypes leaves users stuck at the last mile.

Own the infra to improve agent reliability and shipping speed.

By running agents on their own Kubernetes/container stack and keeping build-time and deploy-time environments consistent, they reduce deployment failures and provide faster feedback loops to agents.

Multi-agent orchestration plus long-term memory compounds capability.

They delegate subtasks (testing, design search, API integration) to sub-agents and aggregate trajectories into cross-session “skills” that are CI/CD-vetted and added to memory, reducing repeated failures over time.

Design quality and UX abstraction drive non-technical adoption.

They hide intimidating developer artifacts (diffs/JSON), ask clarification questions up front, and abstract away setup hurdles like API keys (e.g., “use Emergent LLM key”).

WORDS WORTH SAVING

5 quotes

If you can solve for verification… you can actually automate all the software engineering.

— Mukund

We reimagined the entire platform from ground up saying, ‘What would an end-to-end platform look like?’

— Madhav Jha

Our coding agent is only as good as the feedback that you provide.

— Madhav Jha

Today, eighty percent of users who are on the platform are non-technical users with zero programming knowledge.

— Mukund

In a world of unlimited software, you can build that and seven million other apps.

— Jared Friedman

Founder backstory and early insight (testing as bottleneck)Pivot from testing agents to general coding agentsBenchmark-driven R&D (Suitebench) and multi-agent architectureSecond-mover strategy in fast-moving AI marketsBuilding production-ready infra (Kubernetes, sandboxes, deployment)Non-technical UX design and abstraction of complexity (API keys, code diffs)Personalized/agentic software and implications for SaaS

High quality AI-generated summary created from speaker-labeled 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