After leaving Dunzo in 2023, Mukund spent six months exploring AI tools and building projects without a business plan. That tinkering helped him and his twin brother and co-founder, Madhav, see that coding could be transformed by AI. They chose to pursue autonomous software engineering despite investors' doubts about whether models were ready.
Emergent: How Six Months of Tinkering Led To A $100M ARR Company
Speaker
Mukund, co-founder of Emergent, explains how six months of tinkering after Dunzo led to an AI platform for building and shipping software, in a conversation with a Y Combinator host.
Emergent uses specialized agents for tasks such as design and automated testing, coordinated by a memory system that learns from apps built on the platform. The team developed its own coding agent and infrastructure, including disk and memory snapshots so parallel agents could work from the same state. Mukund says the platform focuses on delivering functional apps with back ends and databases, rather than demos.
At YC, the team tried several directions while working on SWE-bench, a coding benchmark. After three months, its four-person team reached number one, developing techniques including parallel test-time compute, agent memory, and agent-to-agent communication. Mukund says the benchmark gave the team a clear target and helped lay the groundwork for Emergent.
Emergent grew through influencer marketing, which the team modeled as a funnel from social views and impressions to clicks and users. The company is based mainly in Bangalore, with a small San Francisco team; Mukund says 95% of its team is in India. He reports 8.5 million users across 190 countries, over $100 million in annualized run rate, and most revenue from the U.S. and Europe.
Mukund says building for India or for global customers takes similar effort, and advises founders to think globally from the start. He also recommends trusting intuition grounded in customer understanding and setting ambitious goals. From Dunzo, he carried forward the value of solving hard operational problems, staying close to customers, and focusing on what works.
Takeaways
- Use unstructured tinkering to build firsthand insight into emerging technology.
- Set a measurable target to focus a team while the product direction is still evolving.
- Design for the complete customer outcome, not just a convincing first demo.
- Stay close to customers by taking on the hard work of solving their problems.
- Consider global customers from day one when distribution is available online.
Mentioned in this talk
Search inside this talk
What was claimed 19
- metricEmergent: Ranked number one in the world on SWE-bench
“We became number one in the world on SWE-bench”
YouTube Unverified1 source· Talk
- builtEmergent built coding agents
“we started off as a research lab building coding agents”
YouTube Unverified1 source· Talk
- featureEmergent: lets people without programming knowledge build, ship, and monetize software through chatlets people without programming knowledge build, ship, and monetize software through chat
“a platform that allows anybody without any programming knowledge to build software that they can actually ship”
YouTube Unverified1 source· Talk
- founded
“My co-founder, Madhav, is my twin brother.”
YouTube Unverified1 source· Talk
- founded
“My co-founder, Madhav, is my twin brother.”
YouTube Unverified1 source· Talk
- metricEmergent: More than 10 million apps built
“More than 10 million apps have been built.”
YouTube Unverified1 source· Talk
- metricEmergent: More than 8.5 million platform users
“Today, more than 8.5 million people are using the platform.”
YouTube Unverified1 source· Talk
- metricEmergent: Over $100 million annualized run rate
“we're well over $100 million in annualized revenue run rate”
YouTube Unverified1 source· Talk
- metricEmergent: Users in 190 countries
“Today, people are using us in over 190 countries”
YouTube Unverified1 source· Talk
- metric
Dunzo: About 10 million monthly orders at peak“At our peak, we were doing about 10 million monthly orders.”
YouTube Unverified1 source· Talk
- previously at
“I dropped out of the PhD program and joined Google.”
YouTube Unverified1 source· Talk
- role at
“I was a machine learning engineer.”
YouTube Unverified1 source· Talk
- founded
“I started a WhatsApp chat group ... It started with a personal pain point.”
YouTube Unverified1 source· Talk
- metricDunzo: Almost one million riders
“We had almost a million riders on the ground”
YouTube Unverified1 source· Talk
- built
“We built a multi-agent orchestrated system”
YouTube Unverified1 source· Talk
- built
“All of this is coordinated through a large memory system we've built that self-learns.”
YouTube Unverified1 source· Talk
- metricEmergent: Rewrote its system three times in nine months
“In nine months, we've already rewritten our system three times.”
YouTube Unverified1 source· Talk
- usesbenchmark to focus the team and develop coding-agent innovations
“I picked the SWE-bench, the hardest benchmark at the time”
YouTube Unverified1 source· Talk
- metricEmergent: 95% of the team is based in Bangalore
“We have 95% of our team there”
YouTube Unverified1 source· Talk
