Mukund describes how he and Madhav moved from research on software testing agents to general coding agents after realizing that verification could keep agents working longer. Their agents reached the top of SWE-bench within two months. After an enterprise-focused effort proved too slow, they launched a prototype-building pilot in June 2025 and found that most users were non-technical.
AI Is Unlocking Millions Of New Builders
Speaker
Mukund Jha, co-founder of Emergent, discusses building production-ready apps with AI agents and enabling non-technical users to turn business ideas into software on Y Combinator’s The Latent Space, alongside his twin brother Madhav Jha.
Mukund says Emergent aims to help users ship complete products, including backends, testing, deployment, security, and hosting, rather than only front-end prototypes. The company built its own container and Kubernetes infrastructure so the agent can get rapid feedback and build in an environment closer to deployment. Its multi-agent system delegates work such as testing and integration research, while saved trajectories help agents reuse lessons across sessions.
The team hides tools such as code diffs when they intimidate non-technical users, and considers how the agent experiences the platform as well as how the user does. In the demo, Emergent builds a podcast practice app and shows apps made by users, including an audio-video business intake tool and a CRM for lawyers. Mukund says the platform asks clarifying questions and can handle details such as using an Emergent API key.
Mukund argues that coding is only part of taking an app to production, and that users will want to build more as models improve. He says about 20% of apps built on Emergent are agentic, with agents handling workflows inside the apps. He expects longer tasks and teams of agents to require strong verification, including an overseeing agent that checks whether work is complete.
Mukund says many users are small-business owners replacing spreadsheets, email, and WhatsApp workflows with custom apps. He describes EquiMind, an app created by an Alaska-based clinical psychologist and equestrian coach who combined her expertise after finding development too expensive. The founders see software-building tools giving people more control to express ideas without losing details in translation to a developer.
Takeaways
- Build verification into agent workflows so they can work longer and recover from mistakes.
- Treat deployment, testing, and hosting as part of the product when users need working software.
- Use customer conversations to uncover specific workflows people want to automate.
- Hide technical complexity when it gets in the way of users completing their goal.
- Domain experts can build useful niche products when software is affordable and easy to change.
Mentioned in this talk
Search inside this talk
What was claimed 26
- metricEmergent: 7 million apps built in eight months since launch
“in eight months since launch, seven million apps have been built with Emergent”
YouTube Unverified1 source· Talk
- featureEmergent: Platform lets anyone build and ship production-ready software using AI agents
“a platform that lets anyone build and ship production-ready software using AI agents”
YouTube Unverified1 source· Talk
- foundedMukund Jha founded Emergentco-founder
“founders of Emergent”
YouTube Unverified1 source· Talk
- role atMukund Jha is started the deep learning team at
AWS“then started the deep learning team at Amazon”
YouTube Unverified1 source· Talk
- previously atMukund Jha previously at Zenefits
“Maddy went on to Zenefits”
YouTube Unverified1 source· Talk
- previously atMukund Jha previously at
Google“I dropped out of the PhD program, joined Google”
YouTube Unverified1 source· Talk
- previously atMukund Jha previously at Dunzoran a startup in India
“I was running a startup in India called Dunzo”
YouTube Unverified1 source· Talk
- metricSOTA: Number one on SWE-bench within two months
“became number one in the world on SWE-bench within two months”
YouTube Unverified1 source· Talk
- builtcoding agents
“We built coding agents called SOTA”
YouTube Unverified1 source· Talk
- launchedEmergent: Small build-a-prototype pilot
“We launched a very small build-a-prototype pilot in June 2025”
YouTube Unverified1 source· Talk
- metricEmergent: 80% of users are non-technical
“80% of users on the platform are non-technical”
YouTube Unverified1 source· Talk
- metricEmergent: Users in over 190 countries
“they're in over 190 countries right now”
YouTube Unverified1 source· Talk
- featureEmergent: Builds production-ready apps with backend, frontend, deployment, security, and hosting
“replicates what the best engineering teams do: code reviews, automated testing, debugging, deployment, security, and hosting”
YouTube Unverified1 source· Talk
- usesEmergent uses Kubernetesbuilt its own Kubernetes container tech stack
“We built our own Kubernetes tech stack from the ground up”
YouTube Unverified1 source· Talk
- usesfrontend server
“and a React frontend server”
YouTube Unverified1 source· Talk
- usesbackend server
“We have a Python backend server”
YouTube Unverified1 source· Talk
- featureEmergent: Uses multi-agent architecture with delegated agents for testing, design research, and integration research
“delegated tasks go to sub-agents, whether that’s testing, design research, or integration research”
YouTube Unverified1 source· Talk
- featureEmergent: Builds long-term agent memory and skills from prior trajectories through a CI/CD process
“We generate skills based on previous trajectories, run them through a CI/CD process, and add them to long-term memory”
YouTube Unverified1 source· Talk
- featureEmergent: Hides the VS Code editor from non-technical users to reduce intimidation
“we hide it because we’ve noticed that non-technical users panic as soon as they see a diff”
YouTube Unverified1 source· Talk
- featureEmergent: Uses multiple foundation models together to combine their strengths
“we can use multiple foundation models together to extract even more”
YouTube Unverified1 source· Talk
- metricEmergent: Harness extracts 23% more on top of foundation models
“With our harness, we're able to extract 23% more on top of these models”
YouTube Unverified1 source· Talk
- featureEmergent: Supports a mobile app and lets users continue prompting agents on the go
“We also have a mobile app, so you can keep prompting the agent on the go”
YouTube Unverified1 source· Talk
- featureEmergent: Maintains software versions and a release process with a primary owner
“There’s a testing phase and a deployment phase, and we maintain different versions”
YouTube Unverified1 source· Talk
- integrates withCan connect GitHub; used internally for projects
“You can connect GitHub if you want. We connect it internally for our projects”
YouTube Unverified1 source· Talk
- metricEmergent: Roughly 20% of apps built on the platform are agentic
“Today, roughly 20% of them are actually agentic apps”
YouTube Unverified1 source· Talk
- featureEquiMind: Combines clinical psychology insights with equestrian sports coaching
“It combines her insights in psychology with sports coaching”
YouTube Unverified1 source· Talk
