Emergent: How Six Months of Tinkering Led To A $100M ARR Company

Key Points

▶ 0:00 The founder argues that software companies have driven nearly all stock market gains over the last 30 years, motivating Emergent's mission to democratize software creation for billions of people.
▶ 1:14 Emergent is a platform that lets anyone without programming knowledge build, ship, and monetize software; it began as a coding-agent research lab that reached #1 on SWE-bench before pivoting to this vision.
▶ 2:24 Early traction is massive: 8.5 million users, 10 million apps built, and over $100 million in annualized run rate only nine months after the current product launch, driven by high latent demand from non-technical entrepreneurs.
▶ 5:17 Dunzo reached massive scale—about 10 million monthly orders—and became a household name, pioneering the 10-minute delivery trend in India.
▶ 5:39 Core lesson: choose to solve hard problems; despite ~87 competitors, the real challenge was last-mile logistics and delivering products in the right state.
▶ 6:11 Early on, the founder personally made deliveries, embodying “doing things that don’t scale” to gain deep customer closeness and understand the true pain point.
▶ 6:37 The founder is on his fifth startup, not his second, highlighting his extensive repeated startup experience.
▶ 7:26 Watching Steve Jobs launch the first iPhone in 2007 was a defining inspiration, making him want to "bring something to the world."
▶ 8:04 He joined Google's elite search ranking team as the youngest member, which gave him the freedom to challenge the company's then anti-machine-learning approach.
▶ 8:22 The speaker was a machine learning engineer at Google, a role that came with unusual freedom and leeway.
▶ 8:25 They used that autonomy to question existing systems, asking why machine learning wasn’t being applied in certain areas.
▶ 8:32 This initiative directly led them to push some of the biggest changes in Google Search ranking.
▶ 8:35 After a couple of years at Google, the founder got the "startup bug" and left to start a company.
▶ 8:39 His first startup was a group education platform, which raised money but pivoted into a P2P software company.
▶ 8:54 The pivot revealed his real passion was solving education, not the software direction.
▶ 8:57 The founder shut down his education startup and returned investor money because he wanted to build a consumer-first company.
▶ 9:02 He launched a habit-creation startup, got married, and moved back to India when his wife didn't want to relocate to the U.S.
▶ 9:12 Running the startup remotely with his New York engineering team proved extremely difficult, teaching him a hard lesson about remote coordination.
▶ 9:21 The speaker made a deliberate decision to give something up at that time.
▶ 9:25 A key principle from that early experience has stayed with them ever since.
▶ 9:30 They now realize they should do more of one thing: simply trust their intuition.
▶ 9:35 Dunzo began from the founder's personal pain point after moving to Bangalore—managing errands like car service, electricity, and gas led him to think, "there must be an easier way to do this."
▶ 9:51 The first concrete step was creating a WhatsApp group for friends, an informal service where they could ping for help with any task—not a large business plan.
▶ 10:01 The core motivation was to "make life more convenient in urban cities," and solving personal pain points consistently produced stronger user feedback, deepening his connection to the problem and customer.
▶ 10:34 Dunzo was described as a "huge deal" because the founder scaled the company to massive size.
▶ 10:44 The founder shared concrete scale metrics: ~1 million riders, 10 million monthly orders, and ~5,000 stores.
▶ 10:52 The founder acknowledged that this represented "pretty large scale," underscoring the magnitude of operations.
▶ 10:53 The interviewer asks what lessons the founder took from scaling Dunzo, both positive and what he'd do differently.
▶ 11:08 First key lesson: solve a genuinely hard problem — Dunzo faced 87 competitors doing the same thing, making differentiation and success especially difficult.
▶ 11:29 Second key lesson: deep, practical customer obsession — in the pre-AI era, every engineer dropped their work during evening traffic spikes to personally chat with customers, building a culture of caring across the team.
▶ 11:48 Third key lesson: go the extra mile — Dunzo once put a rider on a plane to deliver a packet to another city, showing the team would do whatever it took for customers.
▶ 11:57 Dunzo built genuine customer love by treating every user as a "single customer" through personal, one-on-one service.
▶ 12:02 A key scaling mistake was lack of focus: the dark store model was working well, but the team split attention across 10 other initiatives, and doubling down on the proven model would have helped greatly.
▶ 12:23 The speaker views Dunzo not as a final destination but as a stepping stone in an ongoing builder mindset of continuous creation and iteration.
▶ 12:47 Running Dunzo was "very, very, very hard," with a dedicated Watchtower team monitoring every order in an environment that felt "almost like a war room" because operational things break often.
▶ 13:03 The founder borrowed this operational mindset from Dunzo and applies it at Emergent, where the team monitors all tasks and software and flags anything that is breaking.
▶ 13:14 He credits much of Emergent's operational rigor directly to lessons learned during the Dunzo years.
▶ 13:37 Despite believing Dunzo was “too big to fail” after raising $200M, the founder left in September 2023 feeling “pretty depressed” and spent the next six months reflecting on what could have been done better.
▶ 14:03 During that low period, the rise of ChatGPT and GPT-4 turned building and coding into his escape, leading him to spend 10–12 hours a day experimenting with new models and tools.
▶ 14:31 He describes this as a “luxury” of six months of pure tinkering on things he genuinely liked, with no particular objective, which ultimately set the path to Emergent.
▶ 14:33 The founder describes this phase as pure tinkering on things they loved, with no objective in mind, which became the foundation for Emergent.
▶ 14:54 Very early on, they became convinced that coding would be "disrupted very quickly" by AI—a realization central to Emergent's direction.
▶ 15:15 All the insights gained during tinkering were applied directly to building Emergent, and the joy-driven approach let them go deeper on problems than otherwise possible.
▶ 15:53 After burning out from building a top Indian company, the founder started casually tinkering with ChatGPT and AI models, rediscovering a childlike, natural curiosity that eventually produced deep firsthand insight into how AI capability would progress.
▶ 16:31 Emergent rejected the VC-friendly “copilot” trend and pitched a “crazy” vision—automating software engineering end-to-end—getting rejected by 10–12 VCs who argued the AI wasn’t ready, but the founder saw the gaps as easily trainable given exponential model progress.
▶ 17:04 The guiding principle became a massive, long-term bet: AI progress is exponential, so the company would always build in that direction, focusing on the full problem of software engineering rather than piecemeal solutions—with that downtime and tinkering energy credited as key to finding this path.
▶ 17:26 "Living at the edge" means working with AI models that aren't yet good enough for the task, but are clearly on a trajectory to get there.
▶ 17:41 Despite VC pushback that "models can't do this," the key insight is projecting model improvement forward and seeing that capability is imminent—"you could see the sparks."
▶ 17:51 Many of the best startup ideas come from things that "aren't quite possible yet," making the gap between current limits and future potential the prime space for opportunity.
▶ 18:07 Emergent chose to build autonomous agents rather than copilots, an uncommon and forward-looking decision at the time.
▶ 18:18 The platform is a multi-agent orchestrated system, with coordinated agents like automated testing and design agents entering at different times.
▶ 18:36 A self-learning memory system coordinates all agents, extracting learnable aspects from each new app to improve the platform—a data flywheel reinforced with reinforcement learning.
▶ 18:56 The team does some fine-tuning, but the majority of what they’ve built is their own in-house infrastructure, including the entire coding agent and its supporting systems.
▶ 19:07 They had to invent much of the deep container edge technology themselves because no one was building it at the time and off-the-shelf solutions didn’t meet their needs.
▶ 19:14 They built custom disk and memory snapshotting mechanisms in order to preserve state and let multiple parallel agents run on the same snapshot.
▶ 19:21 Living on the edge lets startups discover key problems early and forces them to build solutions before the rest of the ecosystem.
▶ 19:33 Emergent’s future lies in running multiple parallel agents that "swarm together" to complete tasks.
▶ 19:44 Every new model class demands discarding prior learnings and reimagining the system—leading to three full rewrites in nine months.
▶ 20:10 AI models couldn't reliably output JSON, and 20–30 other YC companies were all working on the same parsing problem.
▶ 20:25 Emergent strategically skipped the JSON problem, betting next-gen models would solve it, and focused on building the agent instead.
▶ 20:32 Their guiding principle was "living on the edge"—building for what would be possible in the next six months rather than current constraints.
▶ 20:42 The interviewer asks if beating the benchmark is core to the founding story; the speaker strongly agrees.
▶ 20:48 At YC, they initially built testing agents, but the founding path was nonlinear — they "stumble upon a different idea."
▶ 21:07 The whiteboard vision was that soon AI would build web and mobile apps, setting the stage for where their tinkering and benchmark breakthrough converged.
▶ 21:31 The team pivoted weekly with an “idea of the week,” causing frustration and chaos until they decided to focus on a single hard benchmark.
▶ 21:54 Choosing SWE-bench as a focused distraction led to cracking it in three months and becoming world number one, laying Emergent’s foundation.
▶ 22:26 Key lesson: attach to a concrete number or benchmark that shows progress — it focuses direction and provides clear feedback for building a company.
▶ 23:39 Existing platforms were focused on front-end demos, but Emergent's insight was that users want real, working software—competitors were good at starting but bad at finishing, with no real back end or database.
▶ 24:26 Emergent approached the problem by asking how to automate all of software engineering, building almost everything from scratch, and validated this by running prompts across all market platforms—massively outperforming everyone else.
▶ 25:10 The team converted growth into a math problem (views, impressions, clicks, users), concluding that an influencer strategy was the right launch approach, which became their ongoing growth engine.
▶ 25:41 ~95% of the Emergent team is based in Bangalore, with a very small presence and a newly opened office in San Francisco, making the company largely built out of India.
▶ 25:53 Emergent is actively hiring; interested candidates can apply directly by emailing mukund@emergent.sh.
▶ 26:03 The founder emphasizes a hiring philosophy centered on "learning slope" and passion for the work, rather than only prior experience.
▶ 26:09 Emergent hires people who are truly passionate about solving difficult problems.
▶ 26:17 The key differentiator is that everyone in the company generally enjoys solving and working with AI, not just one team.
▶ 26:30 The team's motivation comes from the complexity and creative possibilities of day-to-day AI problem-solving, beyond growth or user impact.
▶ 26:38 The interviewer frames the founder's journey as building two very different companies: Dunzo, part of the first wave of hyper-local Indian startups, versus Emergent, part of the second wave of AI-native companies post-ChatGPT.
▶ 26:58 The interviewer asks the founder for key takeaways from building both a first-wave local startup (Dunzo) and a second-wave AI-native company (Emergent).
▶ 27:05 The interviewer asks what advice the founder would give to the audience on where to look for startup ideas and what kind of things to build.
▶ 27:12 Building a local (India-only) company and a global company require exactly the same effort — there is no extra penalty for aiming bigger.
▶ 27:28 Founders should "think global from day one," because local and global startups are equally hard to build.
▶ 27:46 Technology is a great leveler: with internet access available to everyone, founders can reach global customers from day zero, so it's wiser to build for the global market from the start.
▶ 28:00 Follow your intuition — founders should trust their own sense of what customers need over outside advice.
▶ 28:24 Think big — scale your ambition by 10x or 100x, especially in the fast-changing AI landscape.
▶ 28:35 Attack the ceiling, not the floor — the bigger you think, the higher your probability of success.
▶ 28:41 The interviewer thanks Mukund, calling him "an inspiration" and affirming his final advice as "amazing."
▶ 28:49 Mukund thanks the host and audience, describing the energy as "electric" and expressing optimism for seeing "more gigas and emergent coming out of India over the next year."
▶ 29:00 The conversation ends with applause and a final "Cheers," closing the interview.

Emergent: How Six Months of Tinkering Led To A $100M ARR Company

Summary

Executive Summary

Key Points

Video Sections

Exact Transcript