Emergent Says AI App Builder Reached $100M ARR in Nine Months

Mukund Jha describes Emergent as a platform for people with no programming knowledge to build software they can actually ship, have users use, and monetize. The product works through chat with an AI agent, while Emergent handles hosting, deployment, and maintenance. His framing is not that AI makes programmers faster, but that it can move software creation out of the exclusive domain of programmers.

The company’s scale, according to Jha, is unusually large for a product whose current version launched about nine months earlier: roughly 8.5 million users, users in 190 countries, more than 10 million apps built, and more than $100 million in annualized revenue run rate. He said most revenue comes from the U.S. and Europe, with India accounting for about 10%.

The underlying thesis came from a broader observation about economic value. Jha said that over the past 30 years, much of the world’s economic gain has come from software companies; remove software companies from the Nasdaq and S&P, he argued, and the line looks flat. Emergent’s mission is to make that power available to far more people. “There are a billion people with so many ideas,” he said, and many of those ideas die because the people who have them cannot bring them to life.

The users he emphasized are not only hobbyists. Many are entrepreneurs who lack technical teams and have been “handicapped by access to technology.” Emergent’s promise to them is not a prototype generator but a route from idea to working product.

That distinction — shipping real software rather than making demos — became central to how Jha explained the company’s second-mover advantage. He acknowledged that Emergent was not the first AI website builder. When it entered the market, there were already larger players and a number of smaller ones. But the other platforms he saw were mostly focused on front ends and “demo-ware.” In his characterization, they were good at getting started and bad at finishing: users would not get working software, real back ends, or real databases attached.

Emergent approached the market from a different question: if the goal were to automate all of software engineering, what would the system need to do? Jha said the team built “almost everything ground up” and tested prompts across competing platforms. In those internal comparisons, Emergent was “massively outperforming everybody else in the market.” That gave the team confidence to attack a crowded category because, in Jha’s view, the category had not solved the user’s actual expectation: “My software should actually work.”

Jared Friedman framed Jha’s path as an example of what Y Combinator calls “living at the edge”: working where a technology is not quite ready, but where its direction is visible. Jha agreed with that characterization. The models were not yet reliably able to write code, and investors pushed back on the idea that AI could automate software engineering. But Jha said he could “see the sparks.”

After leaving Dunzo in September 2023, Jha spent roughly six months depressed and reflective, thinking through what could have gone differently. AI was accelerating at the same time: ChatGPT had taken off, GPT-4 had come out, voice models were emerging, and new open-source models were appearing. Coding became his escape. He described spending 10 or 12 hours at a time tinkering on his computer, building things with no clear objective.

One project was a Mac assistant that could talk to him. The point was not that this particular prototype became Emergent. It was that prolonged tinkering gave him a deeper feel for where large language models were heading and which categories would be disrupted quickly. Coding, he concluded, was one of them.

That view led Emergent away from the fashionable product category of the moment. Jha said most companies were then building copilots, because that was what venture capitalists wanted to hear. Emergent instead pitched investors on automating software engineering. He said 10 or 12 VCs mostly rejected the idea, even though he was coming out of Dunzo, because they thought AI was not ready.

Jha’s answer was to build for where the models would be in six months, not where they were at that moment. He gave a concrete example: early on, a major pain point was that models could not reliably produce good JSON output. He said 20 or 30 YC companies were working on JSON parsing. Emergent decided to skip the problem, on the assumption that the next models would solve it, and instead focused on agents.

Friedman treated this as more than a company anecdote. He argued that many strong startup ideas come from capabilities that are “not quite possible yet.” Jha’s version was more operational: if a team is close enough to the frontier, it sees the next set of problems before the ecosystem names them, and often has to solve them itself.

Mukund Jha said Emergent began as a research lab building coding agents. Before the consumer product, a four-person team became world number one on SWE-bench, which he described as the benchmark for coding agents. That effort was not originally the startup idea. When the team entered YC, they were building testing agents. They had a whiteboard vision that AI would soon be able to build web apps and mobile apps, but their YC partner suggested the consumer app-building idea might be too ambitious and that they should consider enterprise.

For three months, the company pivoted weekly. One week’s idea might be “AI Zapier”; the team would tinker with it, then move on. The team became frustrated by the constant changes. To give them a hard direction while he figured out what to build, Jha pointed them at SWE-bench, then the hardest benchmark in the space. It took three months to crack, and that work became the foundation of Emergent.

The benchmark effort forced the team to develop techniques Jha later described as core to the product: parallelized test-time compute, memory, agent-to-agent communication, and other pieces of a stronger coding agent. His broader lesson was that attaching the team to a number can be useful when the path is uncertain. A benchmark gave them a feedback mechanism and a way to measure progress toward a hard technical goal.

The current system, as Jha described it, is a multi-agent orchestrated architecture. Different agents enter at different points in the process: a design agent designs the app; an automated testing agent tests it; other agents perform other actions. These agents are coordinated through a large memory system. Jha said the memory is self-learning: each time a new app is built on Emergent, the agents extract learnable aspects and store them, so every new app improves the platform.

Much of the company’s energy has gone into collecting data, running reinforcement learning on top of it, doing some fine-tuning, and building its own infrastructure. Emergent built its own coding agents and much of the system beneath them. Container technology was one example. At the time, Jha said, nobody was building the kind of deep container technology they needed. Because Emergent wanted multiple parallel agents running on the same snapshot while preserving state, it had to invent disk snapshotting, memory snapshotting, and related infrastructure.

The architecture keeps changing with the models. Whenever a new class of model appears, Jha said, the team has to “delete whatever you have learned so far” and reimagine the system through the lens of the new model. In nine months, Emergent had already rewritten its system three times. His point was that the company has had to keep rebuilding around what new models make possible.

Before Emergent, Mukund Jha co-founded Dunzo, which Friedman described as a major Indian startup that raised roughly half a billion dollars. At its peak, Jha said, Dunzo was one of the most loved consumer brands in India and had become close to a verb: people would say “Dunzo it.” The company reached about 10 million monthly orders, nearly a million riders on the ground, and about 5,000 stores overall. It was also, he said, one of the early companies behind India’s quick-commerce and 10-minute delivery trend.

Jha’s account of Dunzo emphasized that the easy part was the interface. In the beginning, customers could simply WhatsApp the company, making it look like a concierge service. That also meant many companies could try the same model. He said 87 companies were doing “exactly the same thing.” The harder problem was the last mile: making sure the product actually reached the consumer in the right state.

His early answer was direct labor. He said he would get night orders, jump on his bike or into his car, and deliver them himself. That was how he connected Dunzo to one of YC’s standard principles: doing things that do not scale. For Jha, doing the work personally helped him understand the customer’s pain point and whether there was real value.

The same operating instinct showed up later in Dunzo’s customer culture. Before AI, every chat had to be manual. During evening spikes, engineers would stop their normal work, return to the chat screen, and talk to customers. He also recalled a customer who wanted to send something to another city; Dunzo put one of its riders on a plane to deliver the package. Jha used those examples to explain how Dunzo earned customer affection: the company repeatedly went “the extra mile” for individual users.

The management transfer from Dunzo to Emergent is not the delivery network; it is the habit of watching whether the user’s requested outcome actually happens. Dunzo had a “watch tower” team monitoring every order, almost like a continuous war room, because operational systems broke often. Emergent applies a software version of that discipline. The company monitors the tasks and software being built on its platform; if something breaks, it flags the issue. Even though Dunzo was an atoms-based business and Emergent is a software platform, Jha’s operating model carries over: the product is not done when the request is accepted, but when the customer receives the result in usable condition.

Mukund Jha was explicit that Dunzo’s ending was “bittersweet.” At one point the company felt too big to fail. It had recently raised $200 million, and he told his co-founder he thought they had reached that point. “Of course,” he said, “it didn’t end that way.”

The main strategic lesson he drew was focus. Dunzo’s dark-store model was working well, but the company was doing many other things at the same time: a marketplace model, pickup and drop, and other lines of business. In hindsight, recognizing that one model was working and doubling down on it would have helped significantly.

That lesson matters because Emergent is also operating in a broad problem space. “Automate software engineering” can expand in many directions. Jha did not say Emergent has solved the focus problem once and for all, but his Dunzo reflection clarifies how he thinks about one risk: the danger is not only choosing the wrong problem, but letting a working core get diluted by too many adjacent possibilities.

His personal background reinforces why this is hard for him. Jha described himself and his twin brother and co-founder Madhav as “idea guys” who constantly have thousands of ideas. With Emergent, the original impulse was to automate more of those ideas into life. Dunzo taught him that ambition has to be paired with concentration.

Mukund Jha described a career shaped less by a single plan than by repeated attempts to build around problems he personally understood. He grew up in a middle-class or upper-middle-class family with an engineer father, studied engineering, and developed an early desire to build something of his own. Steve Jobs was an early influence, especially the 2007 iPhone launch; during a 2008 internship in Spain, he bought an iPhone even though it did not work in India, brought it back as a souvenir, and tried to hack it to work.

In 2009, he went to the U.S. for a PhD, interned at Google, and then dropped out after concluding that Google had already done the research he planned to pursue. He joined Google’s search ranking team, which he said was a 50-person group controlling Google search ranking. As the youngest person on the team and a machine-learning engineer at a time when he said Google was anti-machine-learning in search, he had room to challenge assumptions. He said he eventually pushed some of the biggest search-ranking changes during his years there.

After Google, he started a group education platform, raised money, pivoted into B2B software, and realized it was not what he wanted to build. He returned the money and shut the company down. He then started a habit-creation company, but after moving back to India while the engineering team remained in New York, he found the coordination too hard at that time and gave it up.

The pattern that mattered to him was not the résumé sequence. It was intuition, especially around a problem he felt personally. Dunzo began with the inconvenience he felt after moving to Bangalore: servicing a car, setting up electricity and gas, and handling the many frictions of urban life. He started a WhatsApp group, gave the number to friends, and told them to ping if they needed anything done. Emergent followed the same pattern. He and Madhav wanted to bring more of their own ideas to life, so they started automating programming.

Jha’s advice to founders was to trust that intuition more than the advice they will inevitably receive. His reason was not that advice is useless, but that founders may have a better feel for what their own customers want and need when they are close to the problem.

Mukund Jha said one question had been with him since he returned to India in 2014 after working at Google in the U.S.: why was there no Google or Facebook from India? He pointed to the amount of Indian engineering talent and to Indian leaders at companies such as Microsoft and Google. His ambition after Dunzo was to build a technology-first global company from India.

Emergent is his answer to that question. The company is mostly based in Bangalore, with 95% of the team there and a small team in San Francisco, where it recently opened an office. Jha said the company hires for “learning slope”: people who are passionate about solving problems and excited by the complexity and possibilities of AI. He argued that one thing separating Emergent from similar companies is that people there genuinely enjoy working with AI day to day, beyond the external validation of growth.

His broader claim was that building a local Indian company and building a global company require “exactly same effort.” Both are hard. Because the effort is comparable, he advises founders in India to think global from day one. The internet gives access and reach; technology acts as a leveler; global customers can be reached from India “from day zero.”

He connected this to ambition. Harder ideas can be easier in one respect, he argued, because they inspire stronger people and stronger commitment. For founders considering what to build, his advice was to expand the scale of the idea: “10x that, 100x that.” In an AI period where many things are changing, he said, “it’s not a time to attack the floor, it’s the time to attack the ceiling.”