Cast AI

My co-founder and business partner at Cast AI, Leon Kuperman, wrote a piece on why technology executives should not give up control of AI coding. I agree with every part of it, and I am sharing the argument here. AI coding is becoming the software factory: the production system through which software gets planned, written, tested, reviewed, and shipped. We already know how to govern strategic infrastructure. Nobody casually surrenders control of source repositories, identity, or CI/CD. The system that now operates across all of them deserves the same treatment. The experimentation era has ended. Every engineer inventing a personal operating model was necessary while the industry learned what these systems could do, but it is not a way to run an engineering organization. The early adopters generated enough evidence to see the recurring failure modes: cost governance, IP protection, data sovereignty, and token consumption growing much faster than our ability to measure verified delivery. A few of Leon's arguments: - Privacy by policy asks the enterprise to trust. Privacy by architecture removes the need. A zero-data-retention policy is still a policy, and policies get violated by bugs, misconfigurations, and undocumented behavior. The durable protection is enforcing the boundary at the infrastructure level. - Model selection matters less than the harness around it: how work is planned, which model handles each stage, how failures are detected, how results are verified. A weaker model inside a strong system can detect mistakes and recover. A strong model inside a weak harness can still drift, waste tokens, and produce plausible but incorrect work. - Developer productivity is the wrong metric. The right question is how much verified software delivery an organization can produce per unit of expert human attention. - Sovereignty means owning the enforcement path, not running everything in one region. The scarce resource is not a GPU. It is deliverable compute: the right accelerator, memory, power, jurisdiction, and security boundary, available when a workload needs it. Control of that boundary creates control of financial destiny. As Leon puts it: do not outsource the factory and pretend you still control production. https://lnkd.in/gSrb_hgm

Big news: Today, I am joining the Tokenomics Foundation board as a founding member, and Cast AI is a Founding Premier Member of the foundation, launched today and hosted by The Linux Foundation. The Foundation is vendor-neutral and exists to solve a real problem: there is no standard way to connect what enterprises spend on AI infrastructure to what they actually get back. I've spent years watching enterprises overspend on cloud because nobody measured what they actually used. I built Cast AI to fix that for Kubernetes, and AI infrastructure is now repeating the same pattern with bigger numbers and less visibility. When we analyzed utilization across all our users, average GPU utilization came out around 5%. Most people don't believe that figure until they measure their own environment. Without industry standards, there is no shared baseline, and I wanted Cast AI to help define the measurement. Governance and AI sovereignty are another gap the token factories have not addressed. Enterprises moving AI workloads into production have no standard way to prove where inference runs, which jurisdictions their data crosses, or who can audit the path from prompt to output. Regulated industries end up creating their own controls and taking each AI inference factory's guarantees on faith. If you cannot trace where your tokens are produced or certify compliance on them, then it belongs in the same conversation as ROI. The standards written now will shape how the market measures and buys AI for years. Thank you J.R. Storment for driving this. You turned FinOps from an idea into an industry discipline, and there is nobody better placed to do the same for the economics of AI. Super excited to join the Governing Board alongside this group. Seventeen seats, and every one of them held by someone who has dealt with the economics of infrastructure at scale: Mike Eisenstein, Accenture: Mike Eisenstein Lauren Adelson-Luft, BNY: Lauren A. Sreenivasan "Raj" Rajagopal, Broadcom: Sreenivasan Rajagopal Amit Kinha, DoiT: Amit Kinha Bharat Bendre, Equifax: Bharat Bendre Jay Litkey, Flexera: Jay Litkey John Ridd, Greenpixie: John Ridd William Lobig, IBM: Bill Lobig Arvind Joshi, JPMorgan: Arvind Joshi Eugene Chigirinskiy, Lenovo: Eugene Chigirinskiy J.R. Storment, Linux Foundation: J.R. Storment Christine Sarros, Oracle: Christine Sarros Alon Arvatz, PointFive: Alon Arvatz Frederik Pohl, SAP: Frederik Pohl Dinesh Sonawane, ServiceNow: Dinesh Sonawane Shane Cronin, SHI: Shane Cronin The board list is here: https://lnkd.in/gqUUm2YE More here: https://lnkd.in/gExUkhPP

Cast AI's India region is live today. This one is personal. I've spent the last year talking to engineering leaders across India. Platform Engineering teams, SREs, and CTOs are building AI and inference infrastructure at a scale that would impress anyone. What I kept hearing: "We need automation that meets us where we are." That means full platform capabilities. It means local data residency. It means an in-country team that understands the regulatory environment, not a foreign vendor operating remotely. That's what we built. To our customers across India who pushed us on this: thank you. We're here for the long run! Read more: https://lnkd.in/gYNxWvtu

Kimchi Coding, Cast AI's coding agent, just added MiniMax M3, making us the first coding agent to offer it. Here's why that matters: MiniMax M3 scored 59% on SWE-bench Pro, ahead of several leading commercial models. Its Sparse Attention architecture runs at 1/20th the per-token compute at 1M-token context. Decoding is 15x faster than the prior MiniMax generation. That changes the economics of agentic coding. Kimchi's orchestration engine routes each task to the best-fit model by complexity and cost. Frontier models handle the hardest evaluations. Open-weight models like MiniMax M3 handle the bulk of the work. The result: 2.5x lower cost versus a commercial-models-only baseline, with matching or better quality on spec match and test pass rates. You're not trading performance for price. You're routing smarter. We also built in hard spend caps at every level, from individual API keys to entire orgs, with automatic termination of runaway agentic loops. Deploy serverless on Cast AI or sovereign on your own AWS, GCP, Azure, or on-prem. Frontier coding. Sane bills. Full data control. We added MiniMax M3 to the recipe. Kimchi Coding just got spicier! More info + early access at https://lnkd.in/du6EcCaN

The newest version of OpsPilot, our AI Agent, marks the end of the era of manual Kubernetes tuning. Engineering teams dedicate thousands of hours each year to manually tuning Kubernetes workloads, which includes writing rules, revisiting configurations, and addressing performance degradation after it occurs. Cast AI's OpsPilot goes beyond automation; it replaces the need for manual tuning entirely. The latest version of our AI agent continuously interprets telemetry and operational signals across the entire stack, acting autonomously without requiring any rules to be written by human operators. This feature is now live. This breakthrough is one of the most exciting developments since we started Cast AI, and I am proud of the team that made it a reality. Read more: https://lnkd.in/djjrm_kj

Sam Altman called doubts about AI spending and AI waste "the most fair criticism right now." I'd go further: we have the data. Cast AI analyzed tens of thousands of clusters for our 2026 State of Kubernetes Optimization Report. Average GPU utilization: 5%. The same companies panic-buying H100s are running their existing hardware at 5% capacity. FOMO is expensive. The winners won't be the ones who spent the most. They'll be the ones who wasted the least. Article here: https://lnkd.in/dHeMQQCz

G2 reviews are valuable because they come from real users, not press releases. This quarter, 190 customers independently ranked Cast AI among the top 3 Cloud Management Platforms, achieving record badges and our strongest G2 performance to date. This recognition is a testament to how our product has evolved over the years. A heartfelt thank you to every customer who took the time to review us. Your feedback is invaluable.

Excited to be at AWS Summit Seoul for the very first time as part of Cast AI ’s continued expansion across APAC 🇰🇷☁️ Come visit us at Booth B15 and see how we’re helping teams automate Kubernetes operations, optimize cloud costs, and run AI & cloud-native workloads more efficiently — without the operational headache. What you can expect: ⚡ Live demos ⚡ Kubernetes & AI infrastructure conversations ⚡ Cloud cost optimization insights ⚡ Swag, arcade vibes, and great discussions with the team If you’re attending AWS Summit Seoul, stop by and say hello 👋 See you at Booth B15! Jisu Park | JongJin Lim | Aditya Nanda | Amit Pahwa | Laurent Gil | Ferréol "Fé" Hoppenot | Surbhi Gupta

🔈 Cast AI 𝘼𝙣𝙣𝙤𝙪𝙣𝙘𝙚𝙢𝙚𝙣𝙩 - New EMEA Sales Expert 🌍 👨🚀 - Welcome Back We are so excited to welcome Nikita Redkin to our EMEA team to support our growth in the region. 🚀 Thank you for bringing your passion, belief and strong grit - Let's share our value proposition to all #Kubernetes users in EMEA and support them in their Automation & Optimization Journey. Welcome back to Cast AI - We are fortunate to having you join our team EMEA - Aurélie Ioana Marta Veronika Zoya Amin Etay Guillaume Kim-Norman Sahm Marko Matt Jerome Raphael Skander Tudor Younes Vladimir Sylvain and I look forward to supporting you in this mission. ( 💙💪🚀 ) 🔄 #Kubernetes #APA #AI #Automation #GPU #Cloud #DevOps #SRE #Reliability