Priya Narasimhan

Thank Dani Cherkassky Great piece on the architecture shift voice AI needs. The core thesis: cloud-centric voice systems fail in real-world conditions because they are too slow, too expensive to keep always-on, and deaf to spatial context. The solution is a hybrid architecture—fast, always-on edge processing (Spatial Hearing AI and Cognition AI) handling 80% of interactions locally, with cloud LLMs reserved for complex reasoning. But the Kahneman thesis runs deeper than the article suggests. System 1 and System 2 are not just useful metaphors—they reflect an actual biological sequence. Fast, intuitive processing (System 1) evolved first; slow, deliberate reasoning (System 2) was layered on top. The brain's architecture reflects this: spatial-acoustic processing happens in milliseconds at the brainstem and primary auditory cortex, long before language centers engage. Kardome is not just borrowing Kahneman as marketing. Their architecture is neurologically correct—it replicates the hierarchical, parallel, spatially-aware processing that makes human auditory cognition so robust. Here is the deeper point: hearing preceded language in human evolution, and spatial hearing preceded both. The mammalian auditory system evolved to detect predators and locate prey—survival functions that required sub-second latency and continuous environmental awareness. Language processing was layered on top of this infrastructure. Current voice AI inverts this evolutionary logic: it starts with language models and treats spatial audio as an afterthought. Kardome restores the natural hierarchy. 🙌 Danny Shapiro, Rory Sutherland

35

2 Comments

Excited to share that we’re leading Inception’s $50M seed round, led by Menlo Ventures. Inception is taking a fundamentally different approach to AI — applying diffusion models (the technology behind image generation systems like Stable Diffusion) to code and text generation. While most AI companies are iterating on autoregressive models that generate text token by token, Inception founders Stefano Ermon (Stanford professor and diffusion model pioneer), Aditya Grover and Volodymyr Kuleshov are building models that refine outputs through iterative refinement. This architectural difference unlocks real advantages: Their Mercury model achieves over 1,000 tokens per second — significantly faster than traditional approaches. In a world where compute costs and latency are critical bottlenecks, this parallel processing capability is a game-changer. As codebases grow more complex and AI infrastructure costs become clearer, diffusion-based approaches offer meaningful advantages in processing large volumes of code and managing data constraints. Thrilled to back Stefano and the Inception team alongside Mayfield, Innovation Endeavors, M12, Snowflake Ventures, Databricks Ventures, NVentures, and angels Andrew Ng and Andrej Karpathy. https://lnkd.in/geRVjWRA #AI #VentureCapital #DeveloperTools

209

10 Comments

ERIC SCHMIDT'S AI WARNING IS AIMED DIRECTLY AT SCHOOLS “The arrival of this intelligence—AI, then AGI, and then superintelligence—is the most important thing that’s going to happen in five hundred, maybe a thousand years in human society, and it’s happening in our lifetime. So don’t screw it up.” My Forbes column this month examines what this means for both K-12 and higher education. In it, I explore: => The difference between AI and AGI (important to understand) => The impact of humanoid robots with AI brains arriving in our homes and schools. => Three questions school boards and school administrators need to begin tackling now => Schmidt's final question: “What happens when every single one of us has the equivalent of the smartest human [working] on every problem in our pocket?" https://lnkd.in/ggM6uUnM

9

1 Comment

Productivity is not impact. AI is collapsing the cost of writing code. Everyone’s celebrating. I think they’re looking at the wrong layer. I wrote a paper on what I’m calling The Hourglass Inversion. The short version: → Engineering value lives in three layers. The top: vision, formal verification, constraint-based architecture. The middle: implementation. The bottom: hardware, physics, cache lines, disk I/O semantics. → AI automates the middle. The middle was already the least differentiating part. → The systems that endure were built by people who understood the problem deeply enough to specify it (top) and the machine intimately enough to realize it (bottom). The code was incidental. → As the middle compresses, the top and bottom become load-bearing. → The hourglass has inverted. Three systems, three proof points: FoundationDB — Dave Rosenthal, Dave Scherer, Nick Lavezzo spent 18 months building a simulation framework before their database touched a disk. Deleted Zookeeper after finding two bugs via fault injection. Replaced it with Paxos in Flow. Zero production bugs since. Jepsen declined to test it — their own simulator was more thorough. ScyllaDB — Avi Kivity wrote KVM, then held up an Intel Xeon die and said: “this is a networked device.” Every mutex is a message on the ring interconnect. His response: thread-per-core, shared-nothing, no locks, no GC. 10x Cassandra on the same hardware. Not a better algorithm — a better understanding of the silicon. TigerBeetle — Joran Dirk Greef traced 10-20 SQL round trips per transaction in the Gates Foundation’s Mojaloop switch. All holding row locks. He did the arithmetic: hot rows locked for 1ms = 1,000 TPS ceiling. Horizontal scaling makes it worse. His response: deterministic ordering, Direct I/O, cryptographic checksums at the firmware level. 78 TPS → 2,000+. The pattern across all three: architectural specification preceded implementation. Hardware awareness was a design input, not an optimization pass. Code volume was unremarkable. Impact was categorical. FoundationDB underpins Apple and Snowflake. ScyllaDB replaced Cassandra at scale. TigerBeetle replaced a 600-microservice bottleneck with a single binary. The middle layer is getting cheaper by the month. The top and bottom aren’t. Build less. Constrain more. Measure everything. Paper link in comments. #SystemsEngineering #DistributedSystems #DatabaseDesign #Engineering

25

10+ unicorns. Is it luck or a system? In a recent interview, the conversation shifted (as it often does) to AltaIR Capital and how we keep spotting unicorns so early. I said, "In our latest fund — just over two years old — we already have three unicorns." The interviewer was surprised: "That fast?" I replied, "You know the formula — I get lucky." That phrase has become a bit of a meme. However, when that kind of "luck" repeats over ten times, it's not really luck anymore. I am not a unicorn founder myself. But I've spent years working closely with the founders who built unicorns and watched how even the strongest teams failed due to one simple reason: lack of focus. After thousands of conversations with founders, I started writing down the patterns I kept seeing. That work became the foundation of my new book, Unicorn Focus: The Key to Building Billion-Dollar Startups in the Age of AI, which went live today. I believe that without a few key focus points, success is nearly impossible. In my book, I shared a simple, tactical framework to help founders stay on track: 🔵 Choose one project (not five!). 🔵 ⁠Drive that project with one clear idea. 🔵 Speak to one sharply defined audience. 🔵 Lead with one killer feature and one standout message. 🔵 And communicate all of this consistently: inside the team, to your market, and to investors. Will these tactics guarantee success? No. Will they increase your chances of building something exceptional? Absolutely. Alongside my own stories (both successes and failures) the book brings together insights from founders I've invested in: Andrey Khusid (Miro), Jonathan Siddharth (Turing), Natalie Gordon (Babylist), Mikita Mikado (PandaDoc), Nadav Shoval (OpenWeb), Songe LaRon (SQUIRE) – reflecting lessons we've all learned while building, scaling and backing companies. Over the past month, Unicorn Focus has been Amazon's top new release in Venture Capital, which tells me these ideas are resonating. If you're building a startup right now, focus is your most valuable competitive edge. Grab you copy here: https://a.co/d/3qbjjjl

305

33 Comments

AMI and Yann LeCun are betting the real future of AI is specialized. Their take: world models built on JEPA that reason causally, plan under constraints, and work in messy real-world environments. Their Feb 2026 paper coins it: Superhuman Adaptable Intelligence (SAI). The idea in short is “Outperform humans on specific tasks, tackle domains we simply can’t.” Nabla (clinical AI) is already on board as AMI’s first partner. More architectural innovation instead of more compute. Curious to see where this goes. https://lnkd.in/gmB4k4EA

19

2 Comments

Excited to speak at the Agentic AI conference hosted by #nference. I first came across their work through #MayoClinic, one of our investors, and was struck by their depth in biomedical AI and infrastructure. We’re at a critical juncture in healthcare, where aligning incentives is beyond efficiency, it’s about getting real outcomes for patients. As technology makes problem-solving more tractable, the trillion-dollar question is: how do we ensure those benefits actually reach the patient? Looking forward to the conversation.

17

1 Comment

I’ve been working on the theory of why small, shippable factories are the perfect inversion to the centralized prefab factory model. It all collapsed into one line of math (credit to Gilles Retsin): Factory Efficiency = Usage × (Throughput × Product Value ) / ( CapEx + Deployment OpEx) The idea is that the closer you can drive Usage to 100 % while keeping the denominator tiny, the faster the flywheel spins.  A shippable microfactory tackles this by primarily focusing on a high Usage factor. Because the microfactory can be packed up afterwards and reused, its CapEx gets amortized over multiple projects, effectively raising its overall Usage across the year. If one project alone doesn’t fully occupy the factory, it can simply roll over to another job to maintain high Usage. Its CapEx is often an order of magnitude lower than a centralized factory (< $1M vs. $10-50M min), so the breakeven throughput is more achievable on a small pipeline of work. The real breakthrough here is portability. There is near-zero stranded capital and almost no idle time. To flesh this theory out, I wrote 3000 words on the topic for Brad Hargreaves and Thesis Driven. The article breaks down: - why “fixed factories, shipped goods” is being inverted to shipped factories, fixed goods - what a sub‑$1 M robotic cell does to CapEx per home versus a £45 M off‑site plant - early production/cost data on a shippable microfactory from Mollie Claypool, Gilles Retsin, Sam Baker and the Automated Architecture (AUAR) team - the hybrid reality of robots tackling the heavy structural components and human crews assembling faster and cheaper - technical constraints to pull this model off and future considerations on the optimal business model (own vs. rent) Major shoutout to Gilles Retsin for helping me pull this together. https://lnkd.in/eRTnbZPa

54

2 Comments

🚨 Open-source AI isn’t just a tech issue — it’s a strategic imperative. Clément Delangue (Hugging Face) explains why the U.S. is now making open-source AI a national priority — in response to China’s rapid gains with open models like DeepSeek-R1. But here’s the real question for us in Europe: 🔍 Where is our open-source AI strategy?

2

🗿 Math does not exist to be on a pedestal. It's a rung at the start of a ladder that leads to a broader view. 🪜 🌄 SpeakEZ Technologies was founded to make principled hardware-software co-design a reality - and to solve real problems in the most matter-of-fact way possible. It's all about transparency and efficiency. While others mystify their work with claims about "blurring the lines between P and NP" or "quantum speedups on classical hardware," we believe in being direct. Your performance bottleneck isn't intractable because of some fundamental mathematical formula on a whiteboard - it's slow because it's not the right engineering approach. This isn't magic. And it should be accessible to every business, not hidden behind academic mystique or venture capital buzzwords. Read how intelligent compilation makes "hard" problems practical: https://lnkd.in/dEhEWQPh #HardwareAcceleration #CompilerDesign #FunctionalProgramming #HPC #RealEngineering #fsharp #OCaml #Haskell #Scala #DistributedSystems #SoftwareEngineering

3