Adrian Ionel
Los Altos, California, United States
10K followers
500+ connections
10K followers
500+ connections
About
CEO, co-founder and chairman at Mirantis. Joined Mirantis in 2009 and helped build the…
Activity
10K followers
-
Adrian Ionel reposted thisAdrian Ionel reposted thisA few months ago, I asked Linkedin: “Who’s building AI for audit?" I was tired of dealing with audit firms, and wanted to support disruption in the space Two companies popped up: - One was selling tools to auditors. - The other wanted to replace auditors. Fast forward a few months… - The “replace them” startup pivoted. - Now they’re selling to audit firms too. I get it. It is the safe, logical, revenue-friendly path. But, If your grand vision is to reinvent audit… and your GTM is “sell software to Big 4”… You’re not disrupting anything. You’re becoming the incumbent's feature roadmap. Real platform shifts don’t politely assist incumbents; they compete with them. ✅ Cloud didn’t optimize data centers -- it obsoleted them ✅ Uber didn't sell tools to taxi companies -- it took away their markets ✅ Fintech didn’t just help banks — it became the bank Yes, audit firms own the customers today. Yes, selling tools is easier. But AI changes the cost structure so dramatically that we should at least ask: 👉 Why does AI-native audit even need to look like? What if: - Agentic AI finds all the controls - tests them 24/7 - Evidence is collected automatically - Opinions are generated in real time, At that point… what exactly is the “audit firm”? Let's be more audacious, can we have licensed, automated, outcome-based AI audit services? With no human, or few humans involved? Founders: - Don't sell AI tools to the stack - Be the stack #AI #Startups #Audit #AgenticAI #FutureOfWork #Innovation David B. Cross, Sidra Ahmed Lefort, Daniyal Chawro, Candice (Yi) Luo, Pat Opet, Adrian Ionel, Anand Thangaraju, Grace Cassy, Catherine Hsiao, Catherine Zhou, Hilary Packer, Menny Barzilay,
-
Adrian Ionel reposted thisAdrian Ionel reposted thisEveryone's throwing money at RAG and fine-tuning to fix LLM hallucinations. We went the opposite direction. Instead of adding infrastructure, we added constraints. The result: 22% → 1% hallucination rate with zero training data. Turns out "Explanation is All You Need". Here's what happened: I've spent the past year testing LLMs for enterprise classification tasks; the kind that previously required 6-month ML buildouts at PepsiCo. The flexibility is incredible: just describe what you want, no training pipeline needed. But 20%+ hallucination rates killed production deployment because they required constant supervision. The conventional approach: fine-tune the model, build RAG systems, throw compute at the problem. You end up reintroducing all the overhead you were trying to avoid. We tested whether inference-time constraints alone could match traditional ML reliability. Turns out they can. This new technique; Explanation-Constrained Classification (ECC) Three independent trials, N=891 tasks each: Hallucination rate: 22% → 1% Structural conformance: 100% (with automated validation) Infrastructure required: None Training data required: None Deployment timeline: Days, not months We're now running fully autonomous classification pipelines that previously required supervised ML systems. 6-month projects now complete in days. The insight: You don't need smarter models. You need well-defined problem spaces. This unlocks a category of AI deployments that weren't economically viable before: mission-critical classification where training data doesn't exist or requirements change faster than retraining cycles. Full paper coming soon. Abstract attached.
-
Adrian Ionel reposted thisAdrian Ionel reposted this1,000,000,000 Euro for German Startups - we are finally out of stealth! I am proud to share what I have been working on for the past year. As part of the new #Deutschlandfonds, we are launching Co-Investments in German Scale-ups. After being a fund investor since 2018, KfW Capital is now also supporting startups directly. Until 2030, we will deploy up to 1 billion EUR in German startups at the Growth stage, investing side by side with our portfolio funds and through SPVs. Our focus areas include BioTech, AI, DefenseTech, and DeepTech at Series B+, with ticket sizes of up to EUR 50 million per company. Our first investment is Quantum Systems, one of the fastest growing unicorns in Germany and a critical company for European Sovereignty. We are joining their latest round and are grateful to our partner HV Capital for facilitating this and working closely with us on this investment. Thank you to Dr. Christian Saller and team! More info here: https://lnkd.in/dRBuJqin
-
Adrian Ionel reposted thisAdrian Ionel reposted thisA 7M model just surpassed DeepSeek R1, Gemini 2.5 Pro, and o3-mini on reasoning. It’s called Tiny Recursive Model (TRM) and costs under $500 to train on two H100s for two days. TRM solves ARC-AGI 1 and 2, Sudoku, and Maze tasks. It uses a recursive reasoning loop with five steps: • Drafts an initial answer • Builds a reasoning scratchpad • Compares logic and finds errors • Revises the answer • Repeats up to 16 times It achieves higher reasoning accuracy than models 10,000x larger. Each task costs less than $0.01 to run. Most companies still prompt general LLMs for specific problems. Smaller, task-trained models can now outperform them at lower cost. Architecture, not scale, is driving the next AI shift. 🔗 : https://lnkd.in/g4WqehBp ↓ Are you an AI developer? Check out https://AlphaSignal.ai to get a daily summary of breakthrough models, repos and papers in AI. Read by 200,000+ devs.
-
Adrian Ionel reposted thisAdrian Ionel reposted this“Jeff Bezos … is partly backing a new AI startup called Project Prometheus ….” So named because they’re celebrating stealing AI from the gods or the resulting eternity of torture as a foreshadowing of how the project will end?Jeff Bezos reportedly returns to the trenches as co-CEO of new AI startup, Project Prometheus | TechCrunchJeff Bezos reportedly returns to the trenches as co-CEO of new AI startup, Project Prometheus | TechCrunch
-
Adrian Ionel shared thisSomething is draining 20% of your revenue. You just can't see it. I built Vampire Slayer: it extracts hidden truths from your meetings and finds what's killing momentum, and tells you exactly what to fix. Think McKinsey analysis in 60 seconds. I need 3 business builders to test my latest feature. Takes 2 minutes. Brutally honest feedback required. Here's me using it on myself (yes, I drink my own medicine): https://lnkd.in/gdxibP3c What users say: Satnam Johal (Enterprise SE): "I rarely give 10s. This is a 9.9. When it builds history and tracks progress, it'll be unstoppable." Josh Lowman (CEO): "Oh my god, we totally need this. I didn't even think of this." Doug Romanoff (Managing Partner/VC): "It's magical. You almost don't believe it. Tailored, no-holds-barred business forensics." Want in? Message me. Let's find your vampires. Adrian
-
Adrian Ionel reposted thisAdrian Ionel reposted thisFuck it, here’s my confidential business strategy. At Gold Front, we want to be the best agency in the world. Here’s my secret plan to get there. ↳ 1. Make Gold Front famous. Do this by building my “founder brand.” LinkedIn is the best way I know to drive awareness for Gold Front - and it’s all organic. We spend $0 on paid media and drive 600k impressions a month. ↳ 2. Sell the category. Gold Front is the first and only category design studio. Evangelize this new and better way for founders & CMOs reach their audacious goals. Teach the power of integrated business strategy, positioning and branding by giving away my best thinking and tools for free. And showcase our best work and customer outcomes. ↳ 3. Use increased demand to streamline sales and delivery. Use our marketing to pre-sell clients. A sales prospect who already knows my pitch converts to a paid customer faster and more often than the average prospect. ↳ 4. Go all-in on AI. Human-only expertise is dead. The only viable commercial path is for my shop to be in the business of selling hybrid intelligence - a fluid melding of AI and human expertise. So that’s what I’m building. ↳ 5. Reinvest the profits. The first four items increase our revenue and profitability. Use a grip of these profits to deliver more value to clients - by leveling up both our product and content. And do it all over again. Did I mention I have no idea if this will work? But so far so good. Win or lose, I’ll keep posting photos of me and my dog. Do you know what your company's strategy is? ……
-
Adrian Ionel reposted thisAdrian Ionel reposted thisThere’s one deep insight that drives my LinkedIn growth. Backstory: In order to save my shop, Gold Front, last year I committed to posting every day on LinkedIn. In that time my posts racked up over 600k impressions, I added 4k+ followers, and our pipeline went from empty ($127k) to booming ($1.67m) Here’s the insight behind the success: At a certain point, I just said “fuck it.” I’m gonna be me. I’m going to stop acting like I know everything. Or even a lot. Stop acting like I’m not struggling. Stop acting like I’m not scared. But also stop the false modesty. Stop acting like I’m humble when I’m actually proud. Stop acting like my work doesn’t help launch billion-dollar companies when it does. But most of all, I’m going to stop acting like in order to succeed, I have to be someone else. And so, a bunch of people unfollowed me. It turns out I’m not their cup of tea. (No surprise.) But then a bunch more people followed me, because they like what I’m putting down. And a bunch of you DM’d and said how much my posts mean to you. And CEOs that I advise - that mostly won’t comment on and like these posts - they reached out to say “keep it up.” So something’s resonating. And that resonance saved my shop. So, yeah. “Fuck it.” That’s the insight. That’s the strategy. Weird how sometimes happiness can lead to success - and not the other way around. There’s certainly nothing special about me. You can say “fuck it” too. …….
-
Adrian Ionel posted thisAre you chasing a big, scary goal — but your days feel more grind than fun? My buddy and I are inviting 2–3 people to join a tiny WhatsApp group called “Play in the Storm.” 🌊 Each day, we drop a quick 60-second note — the vampire that drained us, and how we handled it. Win, fail, or somewhere in between. We laugh about our face-plants, and get right back up. It’s fun and honest. You’ll fit if: You’re chasing a concrete goal that truly matters. You’ve got real skin in the game. You want to feel lighter, more energized, and alive every day. Message me here if you want in.
Adrian Ionel commented on a post
1mo
Yes, free, skilled, courageous journalists are among the most important people in our country!
-
Adrian Ionel liked thisAdrian Ionel liked thisWe have 2 engineers in San Francisco and 6 in India. The India team built the foundation of Surface. When we started, the engineering work was split across both teams. SF owned customer conversations and product decisions. India owned the backend architecture. They weren't executing someone else's vision. They were building the actual product. CQRS, event sourcing, domain driven architecture. Real systems thinking. Not quick hacks. What shocked me was the energy. These engineers didn't have to care this much. They could phone it in. Instead they're shipping features end-to-end, thinking about edge cases, refactoring code because they know it matters. One of them was literally the core developer on Astro.js. Another was architecting systems that would normally take a team of five. They were grinding. The reason we could move this fast wasn't just because they were spread out. It was because every single one of them owned what they shipped. No handoffs. No "this is someone else's problem." They built it, they tested it, they shipped it. Everyone talks to customers. SF, India, doesn't matter. When a customer has a problem, the engineer closest to it jumps in. They're not waiting for instructions. They're deciding. They're shipping. Everything combines. The SF team, the India team, the product, the customers. It all works together. We grew fast because we had engineers across two time zones who understood that pushing Surface forward means everyone owns it. That's just a really good engineering team that happened to be in two places.
-
Adrian Ionel liked thisAdrian Ionel liked thisHuge: Unsloth just shrunk the strongest open AI model GLM-5.2 by 84% while keeping ~82% accuracy. This means you can now run a Claude Mythos/Fable class AI on your own local computer 😳 GLM-5.2 from Z dot ai is a 744B parameter open-weights MoE model, with ~40B active parameters per token and a 1M token context window. In plain English, it is massive, it is open, and it is built for long-context coding, reasoning, and agentic work. But the problem is that the full BF16 model is ~1.51TB. So Unsloth compressed it into GGUF versions that are suddenly practical for prosumers: → ~239GB at 2-bit with ~82% accuracy retention → ~217GB at 1-bit with ~76% retention → 4-bit options with ~98% retention → Runnable on 256GB unified memory setups → Someone already got a 1-bit version rendering Flappy Bird, because the internet remains undefeated It’s now clear that the frontier-ish AI is moving from rented infrastructure to owned infrastructure. Your code does not need to leave your machine. Your documents do not need to hit someone else’s cloud. Your agent does not need to stop because an API bill, rate limit, or policy changed overnight. Of course, this is still not everyday laptop territory. You need serious hardware to run it. But the direction is obvious, and AI is now splitting into two worlds: Cloud AI for scale. Local AI for control. And for founders, engineers, researchers, banks, healthcare teams, and anyone working with sensitive data, control is not a feature. It is the whole product. Open-source AI is going to beat the best closed AI models very soon. P.S. check out The Full Guide to GLM-5.2: The ChatGPT Moment for Local AI 🤖: https://lnkd.in/d__8NVYt
-
Adrian Ionel liked thisAdrian Ionel liked thisJohn Jumper leaving Google DeepMind for Anthropic is not a celebrity swap. It is a signal about where the leverage is in frontier AI. The obvious read is "talent war." The better read is this: the best researchers are voting on where the feedback loop is strongest. Not just bigger models, but faster shipping, tighter research to product cycles, and a culture that lets science turn into revenue. I am not arguing that Google isn't doing this right. I am just arguing that Google is too big and there could be frustrations with the bureaucracy. It may even be a cause of Jumper wanting to go to a new place but culture matters to the talent. That matters for buyers too. When a tiny group of labs absorbs most of the top talent, product differentiation starts to look less like feature count and more like platform control, distribution, and trust. In other words, the moat shifts from "who has the smartest demo" to "who can keep the best people, train the best models, and land them in production." The Day 2 question is simple: how much of your AI roadmap depends on a vendor whose main edge may be talent retention rather than durable economics? Again, I want to point out that one shouldn't read much into this move and it may just be a random one off movement but the talent criteria is critical for enterprise evaluation. #AI #Anthropic #DeepMind #EnterpriseAI #TalentStrategy #ModelLabs
-
Adrian Ionel liked thisAdrian Ionel liked thisNutanix ♥️ #SecureEndUserComputing Citrix Omnissa Nerdio IGEL Technology Amazon Web Services (AWS) Azure Virtual Desktop Dell Technologies Lenovo Fujitsu HP AMD Intel NVIDIA Love all our customers/partners! Our team at Nutanix wrote the original book in this market w our partners! Speaking of “original”; California Governor Gavin Newsom and Carl Guardino w/ Silicon Valley Leadership Group announcing the “original” #WYSE #ZeroClient #VDI back in 2010. Great to see my dear friends at Dell Technologies today; they are leading AGAIN #EndToEnd in this amazing market now around #EnterpriseAI 🐝♥️😂
-
Adrian Ionel liked thisAdrian Ionel liked thisMANUALLY DRAGGING BOXES FOR ARCHITECTURE DIAGRAMS IS FINALLY DEAD There is a new open-source agent skill that turns raw codebases into cleanly routed draw.io diagrams without you placing a single coordinate. The project, drawio-skill, runs directly inside Claude Code, Cursor, or Copilot. Instead of opening a blank canvas, you just ask your agent to map the repo. Here is what it actually does: → Extracts the module structure (supports Python, JS/TS, Go, Rust) → Uses Graphviz for auto-layout and routing → Drops redundant edges so the graph stays readable → Builds native, editable draw.io files But the standout feature is visual self-checking. Once it generates the diagram, the agent "looks" at the resulting PNG. If it sees stacked edges or clipped text, it auto-fixes the layout across up to 5 iterative rounds. It runs from a single file. No MCP server. No background daemon. Best part? It's 100% free and open-source. repo link → https://lnkd.in/e6EbXdQS Don't forget to drop a ⭐️!
-
Adrian Ionel liked thisAdrian Ionel liked thisGrateful for the shout-out from Gavin Baker during All-In Podcast Liquidity — and even more grateful for the conviction he has brought to Aria Networks, Inc. from the beginning. The AI infrastructure conversation is shifting from “how many accelerators can I get?” to “how much useful work can I get out of them?” For training, inference, reasoning workloads, and agentic systems, the network is no longer plumbing. It is one of the highest-leverage systems in the AI factory. It affects MFU, token efficiency, cost per token, and the ability to operate reliably at scale. That is exactly why we built Aria Networks #DeepNetworking. To the neo-labs and neoclouds building the next generation of AI compute: we’re open for business. We have supply, a production platform, and a team ready to work side-by-side with operators who want the network to become a competitive advantage, not a bottleneck. https://lnkd.in/gbZ_24JKWhy Secondary Markets Are Eating the IPO | All-In Liquidity Secondary Markets PanelWhy Secondary Markets Are Eating the IPO | All-In Liquidity Secondary Markets Panel
-
Adrian Ionel liked thisAdrian Ionel liked thisAnthropic engineer nailed it: "Delete 90% of your system prompt. You're supposed to use skills for progressive disclosure." That is the exact advice an Anthropic engineer just dropped for fixing broken AI agents. In a recent Code with Claude presentation, Anthropic’s Will Steuk broke down exactly why adding more rules and features is actually making your agents dumber. He exposed the massive gap between how most developers build agents, and how the creators of Claude architect them: → The token bloat that cripples your agent. He showed how to slash a bloated 400-line prompt down to just 15 lines using "Skills" for progressive disclosure: https://lnkd.in/dAg2yXVJ → The custom tool trap. Stop building specialized tools for everything. Give Claude human-like primitives (like Bash execution) so it can write and run its own code to solve problems. A good example is /goal: https://lnkd.in/dXZHxV_w → The sub-agent chaos. Sub-agents cause communication breakdowns. There are only two times you actually need them: throwing parallel compute at a massive problem, or getting a "fresh mind" to review work. → The infrastructure bottleneck. The exact workflows Anthropic's own engineers use to offload the nightmare of scaling and security by deploying to Claude Managed Agents: https://lnkd.in/daSFNGt7 If you're just bolting new capabilities and rules onto your agent every single week, you aren't scaling. You're building a bottleneck. Instead of another Netflix series tonight, watch this talk.
-
Adrian Ionel liked thisAdrian Ionel liked thisA good CRO friend of mine from a F500 makes over 7-figures. Last quarter, his top performing sales rep made MORE than him after bringing in several big accounts. Here's how he responded: CRO: Why is an AE making more than me? Doesn't sit right with me. He's just a rep. Me: Well, he brought in those big accounts. They must've taken at least a year to close and were quite complex. CRO: Doesn't matter. I built this entire sales team from nothing. And it took over 15 years to get it where it's at. If anything, I should get compensated on this deal. Me: You and I get comped differently than ICs. That's just part of being in a leadership role. CRO: Still, this feels rigged. A single rep shouldn't be making over $500K in a quarter. Me: My top AE makes more than me. CRO: He does?! Me: *SHE* does. And she deserves it. She's brought in big accounts, travelled to prospect HQs, wined-and-dined them, gave presentations... she's above quota. CRO: Doesn't that bother you? Me: Not at all. (I'm paraphrasing, but that was the gist of our 15-minute chat). I know someone will disagree with this, but it needs to be said: as sales leaders, we should celebrate when our reps make money. This is what they signed up for. And more importantly, it's what we promised! If your reps start making more money than you, it's a sign you picked the right reps.
-
Adrian Ionel liked thisAdrian Ionel liked thisAI startups should never, ever position against traditional incumbents. That leads to “AI-powered” weaksauce. Instead, position against your strongest AI competitor. (If you can’t find their weakness, you don’t have a strategy.) Here are seven strategies done right: → Superhuman FIGHTS Generic AI BY building email that knows you and learns from you. → Notion FIGHTS Slapped-On AI. BY integrating intelligence right there where your work already lives. → Cohere FIGHTS Unsecure AI BY building AI systems with enterprise-grade security. → Staris FIGHTS Zero Context AI BY by building AI product security with a complete understanding of business context. → Claude FIGHTS AI as a toy BY building foundational AI for builders. → Perplexity FIGHTS Canned AI BY plugging into today's internet instead of delivering yesterday's answers → Llama 4 FIGHTS Closed AI BY building AI that never locks you in. Notice what each of these companies did. They didn't just claim to be better - they named the version of AI they're against. That's the move. …
Experience
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
Education
Languages
-
English
Native or bilingual proficiency
Recommendations received
21 people have recommended Adrian
Join now to viewOther similar profiles
Explore more posts
-
AI Tech Supports
San Francisco-based enterprise software startup Treeline has raised $25 million in a Series A funding round to expand its #AI-powered IT operations platform. The round was led by Andreessen Horowitz (a16z), one of the leading investors in enterprise software and AI. Led by CEO Peter Doyle, Treeline is building an IT Operating System designed to modernise how organisations manage IT, security, and compliance. The platform replaces fragmented, labour-intensive service models with a unified, software-driven architecture. By combining AI with automation, Treeline streamlines routine IT operations while enabling human experts to focus on higher-value areas such as system architecture, risk management, and strategic planning. As businesses face increasing complexity across cloud infrastructure, cybersecurity, and compliance requirements, Treeline positions itself as a centralised solution built for speed, scalability, and security. The new funding will be used to scale the platform, accelerate product innovation, and grow the company’s team as it targets broader enterprise adoption. Hussain Kader Jeff Gaines #Treeline #EnterpriseSoftware #AI #ITOperations #Cybersecurity #StartupFunding #SeriesA #CloudComputing #Automation #Innovation #TechStartup #AITechSupports #Fundraising #AITech #AITechnology #CEO #CTO
-
Salil Deshpande
As AI workloads eat up global computing supply, DRAM prices are surging like never before—creating major challenges for teams trying to plan out their data center spend. One of my portfolio companies, Mext, is addressing the largest cost component in the datacenter: server memory (DRAM). From being on the board of Redis for the last twelve years, I’ve learned a lot about the various aspects of this problem and approaches to solving them. Mext found a breakthrough, using new AI techniques, for dramatically reducing the amount of server-DRAM required to run applications, all while maintaining performance. It intelligently manages the server’s memory, keeping hot pages (i.e., those in use by applications) in fast DRAM and offloading cold pages (i.e., those less used) to a much less expensive memory tier (e.g., NVMe Flash). Key to the approach is ensuring cold memory pages that are about to be accessed by an application are back in DRAM before the application needs them or notices that they were gone. This is done without modifications to the application or the OS – so it can run in the cloud or on-premise. Mext thus allows applications to either run using less DRAM or keep their DRAM footprint but do more with it. They're hosting a webinar along with Fred Weber (former CTO of AMD) on Jan 22nd at 10am pacific time. You can register here: https://lnkd.in/gXe9GCeE
3 Comments -
Jordan Wahbeh
Cloud data-management startup **Eon** raises **$300 million** in a Series D funding round led by **Gil Capital**, boosting its valuation to **$4 billion**—nearly triple from last year. Eon has raised $300 million in a funding round, raising its valuation to $4 billion. The cloud data management startup, founded in January 2024, has now raised $500 million. Eon's software enhances data portability across major cloud platforms. Recent AWS outage underscores demand for robust backup and continuity solutions Increasing enterprise focus on multi-cloud strategies and AI-driven data utilization **Founded:** January 2024; previously valued at $1.4 billion. **Total funding:** $500 million to date. **Investors:** Gil Capital, Sequoia Capital, Lightspeed Venture Partners, Greenoaks. **Growth:** Revenue tripled in the past year; serves “a few tens” of large customers including **SoFi**.
1 Comment
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content