Mukund Jha wanted a game. His father brought home a programming CD instead. For a ten-year-old in late-1990s India, this was a modest betrayal - until the disc revealed a better kind of play. A calculator could be made. So could a game. A computer was not merely a machine that served up somebody else's world; it was a place where he could construct one.
His father was an electrical engineer, and Jha credits him with the original pull toward engineering. Mukund and his twin brother Madhav spent time in internet cafés and chat rooms, played games, experimented with Corel Draw and learned to program. The childhood detail matters because it contains the pattern of Mukund's career: frustration meets a system, curiosity takes the system apart, and then he tries to make it obey a simpler instruction.
School did not supply a tidy founder fable. Jha has recalled bunking classes for movies and even being suspended. He was academically capable but, in his own description, rowdy. At NIT Allahabad, he entered electrical engineering and wanted computer science. A friend in computer science wanted electrical. They swapped. It is hard to design a cleaner metaphor for a technical life: find the mismatch, negotiate with the human on the other side and alter the system.
Five thousand small bets
Jha's early ambition was already pointed at Google. In 2005, he watched the company connect an expanding web through one unreasonably simple box. During college, he worked on speech technology, built a speech synthesizer from scratch in 2007 and spent a 2008 internship at the Polytechnic University of Valencia developing speech recognition. To find research opportunities, he says he sent close to 5,000 emails to universities. The number is less glamorous than a single lucky break and far more instructive. Opportunity was treated as a probability problem.
One lucky seat swap. Roughly 5,000 internship emails. Years of turning improbable outcomes into enough small attempts.
Columbia University took him deeper into machine learning. He published work on natural-language processing and social-data mining, built a content-management system at Columbia Business School and discovered that research alone was not the life he wanted. Madhav continued toward a doctorate. Mukund wanted to make products people could touch.
A 2010 internship at Google put him on an education platform he described as similar to Udacity. He and the team worked nights. A full-time role followed, on Search Quality and ranking in New York, with smaller projects touching news, assistants and the Knowledge Graph. He felt the familiar shock of entering a room full of brilliant people, along with the impostor syndrome that made him want to prove he belonged.
“Something is going wrong somewhere and there are too many variables on how to zero in on what the problem is. That is what makes debugging fun.”Mukund Jha
The useful failures
Google was the dream job. By 2012, the startup world looked more interesting. Jha left and built Wisdom.ly with Columbia classmates, a group-video platform for virtual meetups and conferences. It won a place in Startup Chile, met many founders and failed to find product-market fit. Next came Habet, a 2014 goal-tracking platform that combined expert guidance, accountability and cash rewards. That did not become a lasting business either.
Those misses supplied a distinction that engineers can resist: a product that functions is not necessarily a product people need. Jha returned to India in 2015, tried work in his father's chip-design business, tested ideas in recruitment and then built a WhatsApp shopping assistant with Ankur Aggarwal. They called it Wingman. The instruction was almost comically broad: text anything and get it done.
Blume Ventures introduced them to Kabeer Biswas, who was working on the same problem through Dunzo. In July 2015, they joined forces. Dunzo began as messages sent by people who wanted food, medicine, forgotten keys or a parcel moved across Bengaluru. The customer saw a chat. Behind it, the team had to turn language into tasks, assign supply, understand stores, handle payments and predict what could go wrong.
As CTO, Jha helped move Dunzo from its WhatsApp beginnings to an app driven by automation, data and machine learning. The engineering problem was not a clever screen. It was the thousand-variable mess behind a single request. New user or returning? Known store or unknown? Which partner, which route, which payment state, which promise? Systems were written and rewritten as volume grew. Python gave way to Go where concurrency mattered. Data was cleaned, combined and watched for anomalies.
Dunzo also became something few startups do: a verb. It raised heavily, expanded and then ran into a severe cash crisis. Jha stepped away from daily operations in 2023 and left after eight years as co-founder and CTO. The company later shut down. It was a public ending to a long operating chapter, and an important one. Emergent would not be built by a founder encountering scale for the first time.
A second company, with his first collaborator
The next partnership had been waiting since childhood. Madhav had completed the research path Mukund left, worked on deep learning and helped build Amazon SageMaker. The fraternal twins had talked about starting a company together for years. In 2024, they entered Y Combinator with Emergent as a software-testing automation product.
Testing had bothered Mukund while he managed a large engineering organization at Dunzo. It slowed the release of software. At Emergent, the brothers arrived at a larger insight: if an agent could verify whether software worked, perhaps verification could anchor the automation of the rest of engineering. The testing company widened into a software-creation platform.
“Software development is 20% writing the code. The rest is building it.”Mukund Jha at the India.AI Summit
A user describes an application. Specialized agents clarify the request, design the interface, write frontend and backend code, create databases, connect services, test the result and deploy it. The useful comparison is not an autocomplete inside an editor. Jha calls it an engineering team in a box. The target user is often someone who does not know what an API key is and should not need to.
Company-reported ARR / annualized run-rate figures. Milestones are snapshots, not audited revenue.
The audience surprised the founders. They expected product managers and designers with some technical fluency. Instead they met roofers, factory operators, consultants, logistics companies and property managers. These users carried expertise that software teams lacked. What they did not carry was the budget, time or vocabulary to commission custom applications.
Jha spent Emergent's first five public days glued to customer support. Requests arrived in French and German; AI helped him understand and answer them. It is an appealing reversal. The product existed to translate domain language into code, while its founder used the same technological shift to translate users back to himself.
The growth came quickly. Emergent reported $15 million in annualized recurring revenue within 90 days of public launch, $100 million after eight months and a $120 million annual run rate by July 2026. That month it raised a $130 million Series C led by Creaegis, with MNI Ventures - Claypond Capital and Sentinel Global as co-leads. The round valued the company at $1.5 billion and brought announced funding to $230 million. Emergent said more than 12 million apps had been built on the platform and that it had over 200,000 paying customers.
Where the complexity goes
The numbers are attention-grabbing; the design choice underneath them is more durable. Google hid a web of ranking decisions behind a search box. Dunzo hid a web of logistics behind a message. Emergent wants to hide an engineering workflow behind a conversation. In every case, simplicity at the surface depends on more sophistication below it.
That creates hard tradeoffs. AI-built sites can converge on the same look, a weakness Jha has acknowledged. Reliability matters more when customers are running a business rather than making a demo. Deployment, security, debugging, maintenance and permissions are not decorative extras. They are the tedious remainder inside his 20-percent observation, and the part on which trust rests.
His product philosophy contains a useful restraint too. In AI, some model limitations will disappear within months. A team can spend its scarce time patching every flaw in the present model, only for the next model to erase the problem. Jha has described choosing to let unreliable structured output retry while Emergent worked on orchestration, memory and coordination - problems less likely to be solved from outside. The discipline is to imagine the model six months ahead without confusing imagination for certainty.
Emergent's aspiration now stretches beyond building an app. Wingman - a name resurrected from that pre-Dunzo WhatsApp assistant - is meant to help operate a business across sales, marketing, finance and daily workflows. Jha's public case is that small companies should be able to encode how they actually work rather than squeeze themselves into generic software.
This is where the childhood programming disc returns. The important moment was not learning C++. It was discovering that the machine could be changed by an instruction. Jha has spent the years since trying to shorten the distance between intention and result: from code, to chat, to natural language, to agents working in concert.
The bet is not that expertise disappears. It moves. The engineer's craft goes into the platform; the user's expertise supplies the purpose. A contractor knows the awkward handoff that loses a lead. A factory operator knows where inventory goes missing. A consultant knows which repeated task consumes Friday afternoon. If those people can turn that knowledge into working software, the next useful application may begin with somebody who has never wanted to be a programmer at all.