Accel Atoms

Most founders think the hardest part of selling to an enterprise is building a good enough product. It's not. Recently, Shekhar Kirani, Partner at Accel, sat down with Jason Graefe from Microsoft to talk about what actually separates founders who crack enterprise from those who don't. Here's what it takes to get there and stay there: 1/ 𝐃𝐨𝐧'𝐭 𝐛𝐮𝐢𝐥𝐝 𝐲𝐨𝐮𝐫 𝐜𝐨𝐦𝐩𝐚𝐧𝐲 𝐚𝐫𝐨𝐮𝐧𝐝 𝐨𝐧𝐞 𝐦𝐨𝐝𝐞𝐥. If your product is tied to one model's pricing, capabilities, and roadmap, you're exposed every time a better one ships. The founders building durable companies are treating models as interchangeable so that when models improve, their product improves with them. 2/ 𝐀𝐮𝐝𝐢𝐭 𝐲𝐨𝐮𝐫 𝐚𝐬𝐬𝐮𝐦𝐩𝐭𝐢𝐨𝐧𝐬 𝐛𝐞𝐟𝐨𝐫𝐞 𝐲𝐨𝐮𝐫 𝐜𝐨𝐦𝐩𝐞𝐭𝐢𝐭𝐨𝐫𝐬 𝐝𝐨. Every experienced builder carries a mental map of what's too hard, what's been tried, what doesn't work. AI is quietly invalidating large parts of that map. The question worth asking: which of your constraints are real, and which are just inherited from a world that no longer exists? 3/ 𝐓𝐡𝐞 𝐟𝐢𝐫𝐬𝐭 𝐞𝐧𝐭𝐞𝐫𝐩𝐫𝐢𝐬𝐞 𝐬𝐚𝐥𝐞 𝐢𝐬 𝐚 𝐭𝐫𝐮𝐬𝐭 𝐬𝐚𝐥𝐞, 𝐧𝐨𝐭 𝐚 𝐩𝐫𝐨𝐝𝐮𝐜𝐭 𝐬𝐚𝐥𝐞. Get the first 4-5 customers to build with you, not just buy from you. When it works, they don't just renew; they become the reference that closes your next ten deals. That window is open right now. The founders who define this era won't just be the ones who move fastest. They'll be the ones who move with enough depth that speed alone can't displace them.

Before you scroll, Meesho’s systems have already processed 2 trillion signals to decide what to show you. In an episode of Decoding AI, Anagh Prasad, Investor at Accel, sat down with Debdoot Mukherjee, Chief Data Scientist & Head of AI at Meesho, to unpack what building AI at true Bharat scale actually looks like. Two ideas stood out for anyone building consumer AI in India. 𝐅𝐢𝐫𝐬𝐭: 𝐬𝐞𝐚𝐫𝐜𝐡 A huge chunk of Meesho’s searches don’t look like clean Google-style queries. Someone looking for a fan might type “hawa denewala machine.” Traditional search systems struggle with this because the exact keywords don’t match. LLMs work better because they can understand intent and context, even when the phrasing is unusual. 𝐒𝐞𝐜𝐨𝐧𝐝: 𝐯𝐨𝐢𝐜𝐞. Meesho is already handling 60,000+ voice calls a day, but the interesting part is why voice works so well for Bharat users. Typing vernacular languages on a Latin keyboard is hard, spellings are inconsistent, and keyword search assumes users already know how to “query” a system. The real challenge isn’t just better models, but reducing friction between user intent and the system, whether through LLMs that understand messy natural-language queries or interfaces like voice that remove the need for typed keywords entirely. The companies that win in India will build around how Bharat naturally communicates, not how users are assumed to behave. The full episode is linked below.

Everyone said video was the future. Nobody warned you how hard it would be to actually make one, even when you were doing everything right. When Ritwika Chowdhury, Founder & CEO of Unscript, spoke to a friend running a D2C brand, the problem was immediate: producing a single video took at least two weeks. That meant multiple agencies, internal coordination, endless back-and-forth, and even then, no guarantee the final output would truly feel like the brand. Her first instinct was to solve a specific problem. She built a lip-syncing tool: technically difficult, genuinely useful, and impressive in isolation. But enterprise customers quickly showed her the bigger issue. Even with perfect lip-syncing, teams still had to handle scripting, storyboarding, editing, and brand alignment before a video could actually go live. So Unscript became an AI video agent that handles it all: scripting, storyboarding, editing, and delivering production-ready output end-to-end. And because it learns a brand's style and guidelines over time, it gets faster and more accurate with every video created. Before chasing scale, Ritwika spent the first few months going deep with a handful of customers, not measuring how many new ones she could add, but whether the existing ones kept showing up to create more videos. That was the real test. A growing ARR number can look incredible from the outside, but if customers are quietly churning every month, the growth is fragile. What looks like momentum is sometimes just a leaky bucket with better storytelling around it. For early-stage founders building with AI: don't let early traction trick you. The founders who figure out retention before scale are the ones building something that compounds. Watch the full episode of SeedToScale via the link in the comments.

We raised $3.33 million to build the Optical highway to the Cosmos. Led by Accel in India and Prosus Ventures and given foundational and early support by South Park Commons, ARTPARK and Manish Jain. The entire space industry is racing to launch more satellites. Almost no one is fixing how the data gets back to Earth. That is the gap. QOSMIC is the solution. Here is the problem. More than 100,000 satellites launch this decade. They generate petabytes of data every day. And they still send it down over radio links that have not changed in 100 years. We replaced radio with lasers. QOSMIC is the optical infrastructure. If a satellite needs to move its data to Earth at scale, it comes through us. Here is what that unlocks: → 100x the throughput of RF, at $1 per gigabyte → Argus, Ground stations that catch laser beams straight from orbit → Zaphod, a terminal that puts laser comms on almost any satellite → India's only full stack optical comms platform Built by the team in Bengaluru. We are just getting started. Onward. Shreyaans Jain Rohit Ramakrishnan Aloke Kumar Accel in India Accel Atoms Accel Pratik Agarwal Mahendran Balachandran Prateek Mehta Ankit Chowdhary Aditya Agarwal Jonathan Brebner Akhilesh Agarwal Nayan Nandan Raghu Dharmaraju #OpticalCommunications #LaserComms #SpaceTech #SatelliteCommunications #FSO #DeepTech #SeedRound #Qosmic #SpaceInfrastructure #QOSMIC

Modern satellites generate a lot of data every day. But getting that data back to Earth is becoming a bottleneck. Satellites still mostly talk to the ground over radio. But radio has limits: only so much spectrum to go around, too much traffic competing for it, and short windows to send during each pass. The result is a growing backlog of valuable data waiting to come down. QOSMIC is building the answer. The company is replacing traditional radio links with laser-based optical communications, enabling satellites to transmit data back to Earth at significantly higher speeds and capacities. By developing both optical ground stations and optical communication terminals for satellites, they are building a complete end-to-end optical communications stack. In under a year, founders Shreyaans Jain, Rohit Ramakrishnan, and Aloke Kumar proved their optical communications stack over a 10-kilometre link on the ground. They showed the full chain working outside the lab: pointing the beam, locking on, tracking the target, and moving data at high speed. Mahendran Balachandran • Pratik Agarwal • #AccelFamily

Had a great time on a panel about the startup experience this week, hosted by Jon Silber, Co-Founder and Director of the Google AI Futures Fund, as part of the Google DeepMind Startup Forum organized by Google for Startups. The best part of these conversations is that they cut through the polished version of startup life and get into what building actually looks like day to day. The dead ends, the bets that pay off, the ones that don't, and the decisions you make with far less information than you'd like. It also meant a lot to be there given that the AI Futures Fund backed K-Dense Inc. early. That support came at a point where it mattered, and it has shaped how we build our AI co-scientist platform for researchers around the world. Thanks to Jon and the team for putting founders in a room to share the real lessons. Always learn something when I do.

The barrier to shipping has never been lower. The bar for thinking has never been higher. Deepak Dhanak, Co-Founder and COO of Rocket, says vibe coding has permanently shifted what it means to be a great developer. A year ago, a developer's day was spent translating ideas into code, picking frameworks, wiring boilerplate, and debugging plumbing. Today, AI handles that translation layer. The ones gaining the most leverage aren't the fastest prompters. They're the ones who can describe a problem precisely and hold a sharp definition of what a good product looks like. Read the full piece on Financial Express (India) to know what Deepak thinks the next generation of engineers needs to do differently, and the three things he says they should start doing now. Here's the full piece: https://lnkd.in/g96mXYXK

For more than two decades, we’ve known that dogs can detect cancer from breath. Dognosis makes it real. Last week, their work was published in the Journal of Clinical Oncology, the world's leading cancer journal. For a breath test to clear that bar is a genuine leap for what non-invasive diagnostics can be. The study shows that trained dogs working alongside a Bayesian statistical model can detect cancer across seven major types with 90%+ accuracy, and rule it out with 91%+ accuracy. Most importantly, that performance holds at Stages I and II, when catching cancer early can dramatically improve survival. In India, where rising cancer incidence remains a major challenge, nothing quite like this has existed: a low-cost, repeatable, and non-invasive tool that could be transformative for detection and care at scale. Congratulations to Akash Kulgod and Itamar Bitan on reaching this milestone. Next come real-world programmes across multiple Indian states, followed by a U.S. study. This is LeapTech: what happens when someone chases the improbable long enough that it stops being science fiction. Read the full study here: https://lnkd.in/gQ4nS2ZP 

Researchers don’t spend most of their time actually researching; they spend it reading, trying to map what’s already been said before they can say anything new. Timothy Kassis, Co-Founder & CTO of K-Dense Inc., lived that loop for years and eventually asked a simple question: What if the research process itself is the bottleneck? So he built K-Dense Inc., an AI co-scientist that takes prompts like “draft a PhD proposal,” “map the literature on X,” or “find the gaps in Y,” and turns them into structured, long-form research outputs. It scans 250+ databases, works across 200+ scientific formats, synthesises findings, suggests hypotheses, and produces technical documents that would normally take weeks. The ambition isn’t to replace scientists; it’s to make sure they never have to work alone again. If the hardest part of research is navigating what already exists, this flips the equation. Read the full story on Financial Express (India): https://lnkd.in/gi2pQkDx

Harshit Yadav, Founder & CEO of Zingroll, spent years building consumer apps: an anonymous social network across Ivy League campuses, an AI meme app gaining traction in Bay Area schools. Each one taught him something about what people actually want when they open their phones. What they don't want is 20 minutes of scrolling to find nothing worth watching. That frustration became Zingroll, an AI-native streaming platform built on a simple thesis: every taste group deserves new content every few days, not every few months. Read about the builder, his career-long obsession with consumer habits, and why he thinks 2026 is the year mainstream cinema changes forever on Financial Express (India): https://lnkd.in/gp7Xf67Z