Crane

I've seen enough fundraising announcement videos. This isn’t one of them. At @HacktronAI, we do security, and we do it well. That’s what matters to us. We solve real problems for our customers. On average, they uncover real vulnerabilities missed by other tools within 24 hours of onboarding. Just this year, we've already responsibly disclosed vulnerabilities in Vercel's Next.js, Grafana, Jetbrain's YouTrack, OpenAM, Metabase, and BeyondTrust's Remote Support Software. No unearned, bullshit hype. Just security that works.

I sat down with @ulrikstighansen and @EricEncord, the co-founders of Encord, after their $60M Series C led by Wellington Management. From a wild bet to their incredible success with Physical AI, their story is wild! A seed fund once rejected them because they didn't believe in AI and invested in an Icelandic dating app instead 😅 Here are a few things from the conversation that stuck with me.

Training-ready data for your pipeline. Collection. Curation. Annotation. One partner, end to end. Trusted by 300+ AI teams. Encord is the data layer for Physical AI.

@fence_finance @galaxyhq @paraficapital Congrats team!

Today we announced Fence’s $20M Series A, led by @GalaxyHQ and joined by @paraficapital & @crane_vc We’re using it to rebuild the operating infrastructure for asset-backed finance: the systems that verify assets, enforce facility rules, calculate obligations, and move capital.

Working in tech is a privilege, and the road ahead should be joyful. On June 9, @crane_vc is hosting FESTIVAL: a day at Cavallo Point to step back from the hype and reflect on what we’re actually building. Real dialogue, unfinished ideas, and a focus on the people behind the code. We're excited to hear from @jamesdacombe from @OLIXComputing and @CoMind_, @kenneth0stanley from @LilaSciences, @erishabh from @Pavo_AI and so many more! If this is your kind of room, apply to join us in the link below:

We’re excited to announce the publication of @CoMind_ first two peer-reviewed papers in Neurophotonics.

Most analytical databases start with: • define schema • build ingestion pipelines • create indexes • materialize views • optimize • rinse and repeat The larger the scale, the bigger the pain. What if you just… didn't have to do any of this? We are working on something new.

Two hours with the @crane_vc team went by in a blink, and I left energised. @scott_sage and @krishnav have spent a decade building Crane into a seed-conviction fund for deeply technical founders out of London. Their first fund had a 75%+ seed-to-Series A graduation rate. Now former colleagues from @peakxvpartners - @smdcmc and @Sid_1_0 - have joined hands with Crane to build its India presence through the $150 mn APAC I fund. A serious, structured bet on the subcontinent, and the deeptech thesis. We couldn’t be more excited to collaborate, as Roy is already on our deeptech advisory board. Their plans to back founders building foundational, hard-to-replicate tech for Indian sovereign buyers and global markets aligns with our mission of supporting the deeptech community through a number of programs including 100 DesiDeepechs and Deeptech Baatcheet. Scott, Krishna, @jo_loo, Roy, Sidhant, excited to collaborate on your India journey. And very much taking you up on the offer of supporting SPF’s UK & Europe expansion. 🚀

Tomorrow I'm on stage at PyAI Conference in San Francisco. The lineup is unreal - @gvanrossum, @samuelcolvin, @jlowin, @jxnlco , and @jeffreyhuber, and many many more [pyai.events/speakers]. Some of the sharpest minds in the Python ecosystem, all in one room. My talk: "We Solved Building Agents. Now What?" The short version - everyone's obsessing over which framework to build agents with. But the teams actually shipping are wrestling with something different: the infrastructure that keeps agents alive in production. The harness has two halves, and most of the thinking is pointed at the wrong one. I'm in SF all week. If you're around and want to grab coffee, talk agents, talk infra, talk Python - hit me up. DMs open. See you tomorrow 💪 (Pictured, last time out in SF speaking with Samuel at Modal offices)

Today we’re announcing that @encord_team has raised a $60M Series C led by Wellington Management, with participation from existing and new investors including Y Combinator, CRV, N47, Crane Venture Partners, Harpoon Ventures, Bright Pixel, and Isomer Capital. This brings our total funding to $110M. The world of AI has already changed dramatically since @ulrikstighansen and I started hacking in my parents’ dining room during YC in 2021. But we think the biggest inflection point is actually happening right now. AI is breaking out of the chatbox and into the physical world. Autonomous vehicles are navigating roads. Surgical robots are operating in hospitals. Warehouse systems are making real-time decisions on factory floors. These aren’t prototypes. They’re deployed systems, and their number will grow to tens of millions in the next few years. As these AI systems become multimodal, autonomous, and embedded in critical applications, the quality, governance, and usability of data become make-or-break. These systems are only as good as the data they are fed, and there is very little margin for error. That’s why we’ve built Encord to be the platform AI teams rely on when the stakes are high, to manage, curate, annotate, and align the data their models depend on. Today, more than 300 of the world’s most advanced AI teams trust Encord to train and run their AI on the right data. With this round, we’re going all in on the next frontier of AI because the teams building the most consequential systems in the world deserve the best data infrastructure. We’ve only reached this point through a concoction of hard work, unwavering support, and the shared belief of many people.  To our team, customers, and investors: thank you for getting us here. The hard part is still ahead, and so is the best part. @Wellington_Mgmt @ycombinator @crane_vc @brpxcapital @HarpoonVentures @CRV @N47capital @isomercapital @scott_sage @LarsenJensenUSA @tjrylander

@encord_team just raised $60M Series C led by Wellington Management. $110M total funding to power AI in the physical world. We started in 2021 at my co-founder's parents' dining table during YC, cold outbound by day, hacking by night. Back then everyone was chasing bigger models. We bet that for teams deploying AI in the real world, the differentiator wouldn't be the model. It would be the data behind it. That thesis is now reality. The industry shifted from 'who has the biggest model' to 'who has the right data.' Performance gains now live in post-training, fine-tuning, and evals, and they all start in the data layer. 300+ of the world's top AI teams use Encord today. The next frontier is physical AI. Autonomous vehicles, robotics, embodied intelligence. These systems can't afford to hallucinate. They don't just need more data, they need multimodal, petabyte-scale infra to bridge raw data to production AI. Physical AI is our fastest-growing segment and where we're going all in. To our customers, to Eric and the Encord team across SF and London: we've come a long way from the SoCal dining table. We’re building the data foundation for AI in the real world. We’re just getting started. @Wellington_Mgmt @ycombinator @crane_vc @brpxcapital @HarpoonVentures @CRV @N47capital @isomercapital @scott_sage @LarsenJensenUSA @tjrylander

Book Launch Alert!!!! 📕 On Feb 10, we’re hosting the London book launch of Founder-Led Sales Explained. Most sales books are written for salespeople, but this one is written for builders. rsvp below

Many hardware startups fail by building the device before the soul, the AI. So we build the core of AI first: Visual Memory. Now, the 'brain' is ready for the body. Experience the evolution at CES. We’re giving away 30 Luci Pins for free. Wear it throughout the exhibition and never forget a face, a conversation, or a moment. Come find us and see what total recall feels like. zdnet.com/article/memori…

A founder should go through the full cycle before they even consider changing course. Most people mistake friction for a dead end. They encounter resistance, assume the market has spoken, and abandon ship. In reality, they haven't tested their idea—they have only tested their own shallow effort. To complete a full cycle, you must move past the superficial: Validation: Automated cold emails are not a market signal. High-quality validation requires the manual, unscalable work of forcing a definitive "Yes" or "Hard No" from a real human. Viability: You cannot judge a concept if you wouldn't use the version you built. If you aren't eating your own dog food, you are still in the building stage, not the testing stage. Conviction: If an investor’s skepticism makes you question your direction, you lack ground-level data. True conviction is built on your own discovery, not sideline consensus. Execution is like weightlifting. If your form is bad, switching exercises won't help; you will just carry mediocre habits into the next failure. "This is hard" is not a data point. You only earn the right to move on once you have exhausted every high-quality rep and can explain exactly why the path is a logical impossibility. Finish the cycle. Trust the depth of your work.

We just released a "state of production AI" report of our own! Last week, we shared that our LLMOps database has grown to 1200 entries! This week, we sat down and analyzed each one to find common patterns across the industry going into 2026. This was a lot of work and kudos to Alex Strick van Linschoten for reading through and finding hidden gems. There are some amazing insights here: 1. Real production use-cases have ramped up - Stripe's fraud detection went from 59% to 97% accuracy. Ramp's policy agent handles 65% of expense approvals autonomously. These aren't demos anymore. 2. Context engineering has become a real discipline. Just because you can fit everything into a million-token context window doesn't mean you should. Teams like Manus found that "context rot" starts between 50k-150k tokens regardless of theoretical limits. Leaner contexts make models smarter, not just faster and cheaper. 3. The bottleneck is engineering, not models. The teams winning aren't waiting for GPT-6. They're investing in durable execution, distributed systems, proper evaluation, and treating AI systems with the same rigour they'd apply to any critical infrastructure. This and a lot more were released today on the ZenML blog. Full analysis (with the detailed case studies): zenml.io/blog/what-1200… Executive summary version: zenml.io/blog/the-exper…

High-cardinality metrics break observability budgets - but they shouldn’t ⚡️ └ Traditional systems charge per active time series └ You’re forced to limit cardinality or lose granularity └ High-cardinality becomes a problem, not a capability We built something powerful 👇

Hot off the press 🗞️ Our leap from long-context video understanding to on-device LVMM 2.0 with @Qualcomm is making headlines.

I’m hiring a Chief of Staff to work with me across @flux_computing and @CoMind_ to drive execution and alignment across both teams. We are scaling both companies rapidly - apply below. linkedin.com/jobs/view/4331…