We spent several months working deeply on this investment, and it is a very special moment to finally see it out in the world. When we first met Sujoy and the team, we knew immediately that this was a company we had to champion at Chiratae. Building physical quantum hardware is not an easy path, but their approach made complete sense. Most quantum computers need huge, expensive freezing machines just to stay stable. Quanfluence is building a computer that runs on particles of light instead. By using light, they skip the extreme freezing requirement, which makes the computer much more practical to actually build and scale. They design the physical chips, the control electronics, and the software from the ground up, right here at home. It takes a very rare mix of skills in physics, hardware, and software to pull this off, and this team has it. Leading this investment is a privilege and we firmly believe this founding team will put India at the center of the global quantum landscape. Incredibly proud to partner with Sujoy, Biman, Ravi and the entire Quanfluence team! 🚀🚀 @ChirataeVC @venkypeddi

Some interesting numbers from ICONIQ 2026 State of Scaling: The Great Sorting report! They tracked 137 software companies and split out a cohort it calls 'Pacesetters'. > Median age of cos is six years with 60% AI-native and 70% on hybrid or usage-based pricing. > The chart breaks down where each group’s new revenue in a year actually comes from: brand-new customers, or more spend from customers they already had. > For Pacesetters, over 70% comes from brand-new customers. For everyone else it is about 54% (below $100M of revenue) and 46% for those above it which means the older companies now get most of their growth by selling more to their existing base. > Separately, Pacesetters hold net dollar retention of 120–130% against 99–102% for the rest. So the fast companies are not growing on expansion instead of new logos, they are doing both, and new logos still dominate.

Super super proud of @rokrsa and the whole @tm2space team today as MOI-1a reaches orbit aboard the orbital transfer vehicle on @SpaceX. It gets deployed in the next few days, and from thereon it forever changes how space-data compute happens !! 🚀🚀@ranjith_menon @ChirataeVC

Some journeys you don’t get to be part of, but you get the privilege of watching closely. Watching @venkypeddi and our team celebrate @rentomojo's listing today is one of those moments. I wasn’t around for the RentoMojo investment, but working with Venkatesh has given me a glimpse into the kind of investor he has been for the company - there through the good years, and through the many years of building along the way. This is a special one, watching someone that I work closely with celebrate a journey like this. And, of course, none of this without the vision, grit and years of building by Geetansh. Congratulations Venky, here's to many more! @ChirataeVC

Bangalore rains finally

Samsung’s social media team working overtime today xD #appleevent

Haha, knew something like this was coming!

Excited about live recap and Siri Recap features in the new apple watch, going to be a life-saver! #appleevent

Crazyy! This will have implications far beyond theoretical math. Imagine the breakthroughs in engineering-modelling if we can truly predict turbulent flow.

Spent some time reading about OpenAI's Astra performance. Until now, frontier models were useful assistants to a hacker. You still needed the hacker. Astra is OpenAI's assertion that's no longer the case! On ExploitBench, a test that takes a known software bug and asks whether the model can turn it into a working exploit, Astra scored 100%, against 78.5% for its predecessor GPT-5.6 Sol. In another harder test that OpenAI built specifically to rule out any memorisation, Astra found 20 high-severity bugs in Chrome's JavaScript engine from June to August 2026! (thats too recent to be sitting in training data!!). During that run, it also found two vulnerabilities nobody knew about, which OpenAI says it has reported to the maintainers. A flaw which nobody has discovered yet is a zero-day. Those who follow the world of cybersec know that Zero-days are the scarce commodity of offensive security; a good one in a browser is worth six or seven figures on the open market and takes a skilled human team weeks to months to produce. Astra produced two of these as a side effect of being benchmarked!! Wild to imagine what 2026 is going to end with.

Over the past few months the word 'harness' has become the standard term in the AI world for everything that wraps around a model in an agent product. The idea behind the word is that by late-2026 the harness, not the model will be where most of the performance difference lies. But what exactly does harness mean? Think about hiring a very fast, very capable contractor. This person has no memory of yesterday. They’ve never seen your building...If no one tells them what 'done' means they will confidently do the wrong thing. You could keep hiring contractors. Or you could build the site around them: hand them the plans before they start, give them the tools, place an inspector at the door who checks every wall before it counts as finished and write down every mistake so it cannot happen twice. The plans, the tools, the inspector, the rules make up the harness. The model is the contractor. A popular equation has taken over the industry: Agent = Model + Harness (credits to @Vtrivedy10 and the famous statement 'If you're not the model you're the harness'!) The reason the word caught on is an observation: the same model, in two different harnesses can produce very different results. One team will get a demo that impresses but then breaks in production. Another team may get a fully functioning production-grade software! The difference isn't in the model's weights. This shifts how we see progress over the year. It's not all model progress. A lot of it is harness progress. Harnesses are built with software engineering not just through training runs. However, general agent-harness features are being commoditised at speed by four labs racing each other; a startup whose differentiation is a better loop around a model is standing on ground the labs will take. A startup that has encoded a specific industry's guides e.g. the checks that define 'correct' in claims adjudication, in drug safety reporting, in Indian GST filing, owns something no one else can ship. And while the word catches on further, my view is that in the perennial incumbents-vs-newcomers debate, incumbents armed with massive data moats may find themselves winning this again! :)