Krishna Mehra

Seventeen Years Today @capillarytech goes public. Seventeen years since three kids fresh out of college decided to start something in the wreckage of 2008. I wasn't there for the second half.. the harder half. @AneeshBReddy steered this ship through storms I can only imagine. I'm grateful beyond words. What I remember most isn't the milestones. It's the texture of those early days. Merchants in Kolkata who humored us when we had nothing. Durga Puja 2008, camped outside police headquarters chasing traffic updates. Pan IIT at IIT Madras where our scrappy SMS service caught fire. That first angel check of ₹15 lakhs from IIT Kharagpur that we paid back in crores. We had no idea what we were doing. Just energy, naivety, and this strange refusal to think small. Then Aditya Birla and Future Group said yes. Right after Lehman fell. I still don't know why they trusted us. The apartment in CV Raman Nagar with no water during the day. Chai walks in BTM Layout with @shubhmalhotraa, Abhilash, Prakhar, @pigol1.. trying to figure out if we were building something or just fooling ourselves. Then loyalty took off. That first campaign—₹11 lakhs in incremental sales, one weekend, fully attributed. That's when we saw it. A chance meeting with @KartheeMadasamy at CCD led to the @Qualcomm QPrize. We were shocked when we won. The business grew, but we stayed scrappy. That tiny hotel in Dadar because it was cheapest. The Bangalore home-office. The Gurgaon apartment where Ajay held the fort. Holed up in Bur Dubai searching for product-market fit. The exhilaration when @sequoia (now @peakxvpartners) and @NorwestVP bet on us, alongside angels like @RajanAnandan, Venkat, KS, and Harminder. But what I really remember are the crazy things. Flooding someone in Gujarat with hundreds of thousands of texts. The new hire who nearly wiped our database on day one—later led all of engineering (take a bow, @pigol1). Anant building a POS system on an Excel sheet to close a deal. Sujatha, our first female hire, making sure our office was no longer a dorm room. Building loyalty systems where internet barely worked—the magic of "nightly sync." We were pirates, not professionals. That's what made it magic. Like humans, companies find their early years most thrilling. The teenage years? Brutal. But at seventeen, she's finally ready—scarred, wiser, infinitely stronger. We remember our failures more than our wins. We carry the scar tissue. But today, we celebrate. What a journey, Capillary family. What a journey.

today, we are making the @mosaic_so video editing api available to all agents & humans. see how we setup larry — our slack openclaw agent — with a mosaic api key and had it clip, edit, and post archived steve jobs footage. all without ever leaving slack. you can connect your openclaw agents, claude code, or vibe-coded saas to our agentic video editing api. no waitlist — now live at mosaic [dot] so. comment "API" to get 100 free credits dropped into your account to get started with the api for free. more details on why & how we’re making this change below (thread):

TIL: Stunts and Virality on X don't build a successful business all by themselves, sometimes you need a working product too 🤠

Where is @github CoPilot at these days? Asking for a friend.

@chamath You should use @PortkeyAI for model switching and team based token cost budgeting. @jumbld

Just to add some color to the commentary about Anthropic and Cursor's revenue ramps, from the perspective of a customer. Since November 2025, our AI costs have more than tripled and we are now spending many millions per year trending to $10M+ per year. That, in and of itself, feels very scary to me running a small startup. Mostly because I do not yet see an equivalent uptick in productivity or revenue...while their revenues may be doubling and tripling every month, ours are not so this is starting to eat into margins. So logically, I am now wondering how much of this is models running in Ralph loops on behalf of an engineer ambivalent to how much it costs. My suspicion is a lot! Anyways, we are making a few decisions: 1) We need to migrate off of Cursor. Its just too expensive vs Claude Code. The latter is equivalent and if you use the Pro plan, you eliminate huge Cursor bills for token consumption. Thank you to the VCs who will fund this all you can eat token consumption through their huge investments. 2) We need to gain more flexibility to swap between models without everything breaking. I think this is both a cost problem per #1 but its also a strategic flexibility issue after the events between Anthropic and DoW. Anyways, just sharing our current lived experience as I suspect many other companies are also feeding this revenue ramp without getting any meaningful ROI from it...

@Rahul_J_Mathur Congratulations @Rahul_J_Mathur

On this auspicious day of Holi - I am starting a new chapter by moving into a new home in Indiranagar. I had my eyes on the property since moving to BLR in April 2024 - God is kind - came across the vacancy in December 2025 during an evening walk & closed paperwork in a few hours. Did a Ghar Pravesh puja per request from my grandmother (streamed it for her) 🪔 Happy Holi 🙏

Two years ago, I believed that there was no way AI foundation model companies could “grow into their valuation” just through subscriptions and would need advertising or other mechanisms. How wrong was it!!! With AI, it’s time to revisit your priors.

Is it just me or do others also feel that @ChatGPTapp has basically become unusable in the last few weeks?!. My usage has entirely shifted to @GeminiApp and @claudeai

The world of AI is full of paradoxes right now 🧵 NVDA is down because too much capex in AI. Rest of software is also down because too much capex in AI. Cost of building is almost zero. AI companies are raising more than any cycle before. AI adoption is through the roof. Enterprise folks I talk to still ask me how to set up Claude. AI was supposed to make interfaces intuitive. The killer UI turned out to be the terminal. AI needs high-quality human-made data to get smarter. It's flooding the internet with so much synthetic slop that it's poisoning its own future training data. AI was supposed to make creative work fulfilling. Every engineer is now a "manager of agents." Building product has never been easier. So everyone's chasing influencers. TikTok as the preferred marketing channel for enterprise software was not on my bingo card. Everyone's worried about AI taking jobs. Hardest thing for every AI startup? Hiring. Everybody joked about AI wrappers. Those wrapper companies are decacorns now. AI was supposed to make our jobs easier. Everyone in AI is working harder than ever. Biggest bottleneck for AI growth? The good old power grid. Fastest way to drive AI adoption? In-person meetups and intimate dinners. Weren't the agents supposed to decide everything?

We just raised our $15M Series A to scale our unified control plane for production AI. AI is now mission-critical infrastructure, and we’re building the reliability layer so it never breaks. Thanks to our investors @ElevCap + @lightspeedvp portkey.ai/blog/series-a-…

Huge congratulations to @PortkeyAI on raising their $15M Series A funding! As AI rapidly becomes the core infrastructure for modern enterprises, it brings two massive operational challenges: ensuring reliability at scale and controlling unpredictable, skyrocketing costs. Led by @jumbld Rohit Agarwal & Ayush Garg, Portkey is transforming how enterprises manage production AI and agent operations with its unified control plane. Portkey’s platform layers a high-performance AI production stack with built-in governance, observability, reliability, and cost management. Sitting directly in the path of AI traffic, where tokens become dollars; it ensures engineering teams maintain operational control while giving finance teams real-time visibility into every cent spent. The scale they've achieved already is phenomenal. Portkey is trusted by AI-forward enterprises like Postman and Snorkel AI, processing over 500 billion LLM tokens across 125 million requests every single day. Today, they are managing over $500,000 in daily AI spend for more than 24,000 organizations globally. We are thrilled to lead this round, with participation from Lightspeed. Wishing Rohit, Ayush and the entire Portkey team continued success and many more milestones to come! @kpowerinfinity @PoorviVijay

Very excited to share our partnership with @PortkeyAI as they build the enterprise production stack for the AI era. From the first meeting with @jumbld and Ayush, we have been deeply impressed by their clarity in product strategy, ability to think two-steps ahead, and build what the world will need as AI starts to scale. Their customers told us in droves how much they loved the Portkey’s approach — the fact that it “just works” and is the central choke point that manages and governs their AI strategy. Layering on an enterprise go-to-market motion on incredible open source community love, we believe Portkey is leading an infrastructure category that every organization serious about AI is going to need. Stoked to see Rohit and Ayush building a category defining company and play a part in supporting them. @ElevCap Prerna Gupta Dhruv Jain @PoorviVijay @mukularora

Huge news: our team at Portkey has raised $15M in Series A funding led by @ElevCap w/ @lightspeedvp 🎉 As agentic workflows become a reality and businesses recognize AI as mission-critical infrastructure, we’re here to ensure it never breaks. (1/5)

In the pedestrian world of 4X revenue multiple public software companies, being a pre-revenue neolab unicorn is the ultimate flex!

Great article on software moats and how they are changing with AI. Worth a dekko if you’re worried about SaaSocalypse.

Wednesday night reminded me why I love working in AI right now. We brought together Distinguished Engineers from frontier labs with founders and CTOs building at the edge of what's possible. No panels, no presentations—just honest conversation about what's actually working and what's harder than it looks. A few things that stuck with me: 💬 RL environments are becoming an operations business. Everyone talks about "training in simulation," but the hard part isn't the code—it's designing tasks, building verifiers that actually correlate with real-world performance, and coordinating the humans who still matter. The real moat isn't infrastructure; it's auto-verification that works reliably enough to remove humans from the feedback loop. We're only there in narrow domains: coding (where tests don't lie), structured workflows, rule-bound systems. Everywhere else, you're back to expensive human labeling. 💬 The abstraction layers we take for granted might disappear. One conversation that kept circling back: if you have strong enough verification systems, why do we need programming languages at all? Why not go straight from English to compiler-level code? It sounds wild until you realize we've already retrofitted coding agents to work this way in constrained environments. The question isn't if, it's when and where. 💬 Publishing is dying, and research is splintering. The open-paper culture that defined ML is fading. Important work on JEPA, new architectures, miniature teaching models—it's happening, but in private channels and Discord servers. Berkeley is graduating 1,000-student AI cohorts who've never known the old norms. The talent wave is here, but the knowledge-sharing infrastructure isn't keeping up. 💬 Diffusion LMs might be the next frontier model architecture, but the challenges are real and unsolved. Synthetic data is hitting diminishing returns faster than anyone wants to admit. Multi-agent swarms are already part of researchers' personal workflows—not as a research curiosity, but as a daily tool. 💬 One throwaway line that made me laugh: "There are hacks to get a $40k coffee machine without paying for it." (No, I won't elaborate. You had to be there.) Thanks to everyone who showed up and made it a real conversation. These aren't the discussions that happen on Twitter or at conferences. They happen when smart people stop performing and start thinking out loud together. @ElevCap Dhruv Jain @PoorviVijay @madiator @AlexGDimakis @anuragranj @krishnanvijay @matthewjsargent

@kamath_sutra Kicked to see more here. Two fantastic folks to follow @kamath_sutra and @shortkingceo

Announcing... Voice x Memory! We’re unpacking what makes agents listen, respond, and remember, or sometimes forget, and what that means for building better voice systems. We will move towards a world of large LLMs remembering a lot of information to smaller LMs with finite real-time intelligence and infinite memory. Here's a short teaser as we walk down the streets of SF having a chat with @shortkingceo on how we see the future

All one can say for sure is: You know nothin’, Jon Snow 🐉