Articles by Dan
-
The Rise of the Probabilistic Founder
The Rise of the Probabilistic Founder
Over the last few months, I’ve been having versions of the same conversation with a bunch of seed and early-stage…
32 Comments -
The Token Threshold - Is your Buyer actually an Agent?
The Token Threshold - Is your Buyer actually an Agent?
“The Token Threshold is the point at which the cost to build a capability from scratch exceeds the cost to buy it…
26 Comments -
Swipe Right For A Cloud Instance: How To Build A Successful Software Company In A Low-Commitment World
Swipe Right For A Cloud Instance: How To Build A Successful Software Company In A Low-Commitment World
This post originally appeared in Forbes. Co-authored with Battery Ventures General Partner Dharmesh Thakker.
3 Comments -
Back To The Future: Chip Makers Are Putting The Silicon Back In Silicon Valley
Back To The Future: Chip Makers Are Putting The Silicon Back In Silicon Valley
This post, which I co-authored with Battery Ventures General Partner Dharmesh Thakker, originally appeared in Forbes…
-
The BOSS Index: Tracking the Explosive Growth of Open-Source Software
The BOSS Index: Tracking the Explosive Growth of Open-Source Software
This post, which I co-authored with Battery Ventures General Partner Dharmesh Thakker and former CEO of MongoDB Max…
-
Blue-Collar Revenge: The Rise Of AI Will Create A New Professional Class
Blue-Collar Revenge: The Rise Of AI Will Create A New Professional Class
This post, which I co-authored with Battery Ventures EIR and former CEO of MongoDB Max Schireson, originally appeared…
-
Can viral open-source projects be enterprise powerhouses in the near future?
Can viral open-source projects be enterprise powerhouses in the near future?
This post, which I wrote with Battery Ventures General Partner Dharmesh Thakker, originally appeared on TechCrunch. In…
Activity
9K followers
-
Dan Nguyen-Huu shared thisHad a great time sitting down with Peeyush Upadhyay for The Scaleup Flight pod last week. We covered a lot of ground: → Building and scaling in the AI era. What actually changes when you're building a company now versus five years ago, and what doesn't. Finding real customers with real urgency is still the hard part. → Cybersecurity in a world of agents. Agents acting on your behalf, with your credentials, at machine speed. That's a different threat surface than the security stack was built for, and it only gets more important from here. I believe this will have downstream implications on all security operations. → How we work with founders on product-market fit at Decibel Partners. Getting them in front of practitioners early and often. Early adopters who'll poke at a rough product and tell you what's wrong with it are worth more than any advice we can give from the outside. Thanks again for having me on and for the lively discussion! Episode coming soon.
-
Dan Nguyen-Huu shared thisSo fun meeting Learner Tien last night! If you know me, you know I've been a big tennis fan all my life, having grown up playing in Germany. Michael Chang was my guy back then - poster on the wall, tried to chase down every ball like him, the whole thing. Learner's coached by Chang now, which made it even easier for me to be an instant fan. The sport is in the middle of a real generational transition. Learner turned pro in late 2023 and is already No. 12 in the world with two titles and an Australian Open quarterfinal, at 20 years old. One of the best things he told us: before every match he pulls up a YouTube video of his opponent getting absolutely dismantled. Said it took some searching when he drew Novak last year 😅 . Big thank you to Mats Merkel and Jack Howard for pulling a great group together last night. Rooting for you at the Open, Learner!
-
Dan Nguyen-Huu shared thisHere it is — the full recap of our Game Day event at Hacker Summer Camp! Thanks to everyone who made it our best one yet. Edward Wu David McGuire Andrew Peterson Ashish Rajan 🤴🏾🧔🏾♂️ Zack Allen Jack Naglieri Damien Lewke Kevin Mandia Rob Joyce Casey Ellis Jonathan Cran Zakir Durumeric Elias Manousos Brandon Dixon 🇺🇸 Josh Kamdjou Daniel Miessler Decibel Partners Jon Sakoda Stacey Wueste
-
Dan Nguyen-Huu shared thisGot to do an episode of CyberBytes with my good friend Andrew Peterson while we were all at Black Hat. Andrew co-founded Signal Sciences. We met when he was raising one of his earliest rounds. I unfortunately didn’t get to invest at the time, but we’ve been co-investing together ever since in companies like Sublime Security, Nebulock Fixify Mallory Dreadnode SpecterOps and many others over the years. Oliver Legg asked what makes a founder a “hell yes.” Andrew said team-market fit, which I think is 100% right. Mine was just slightly simpler. How willing are you to get on the plane? In the era of AI, building trust with customers and showing up to help in person has never been more important. Full episode is out now. Watch or listen where you find your podcasts. Thanks again to Oliver and the Aspiron Search team for having us on the show! PS: Good lesson for me personally that white pants are not the way to go.
-
Dan Nguyen-Huu reposted thisDan Nguyen-Huu reposted thisIf you're an early-stage founder wanting to raise capital, or a security investor, this is not an episode to miss.... Andrew Peterson, Co-founder and GP of Aviso Ventures and Dan Nguyen-Huu, Partner at Decibel Partners joined me on CyberBytes: The Story at BlackHat a couple of weeks ago. This was a great episode, partly due to the vibe and relationship these two have as friends and regular co-investors, but also because of the actionable insights and advise they gave on how they approach their investment thesis and how early-stage founders should work with and approach raising capital. 📺 YouTube & 🎙️ Spotify links in the comments, enjoy the show!
-
Dan Nguyen-Huu shared thisWoke up to a nice surprise in my weekly reading: a shoutout in Detection Engineering Weekly. Thanks Zack Allen 🙏! DEW is one of the newsletters I read every week and have been a fan of for a very long time. If you care about detection engineering, threat hunting, or just want a well-curated read on what attackers are actually doing right now, it's one of the best out there. Was great to get Zack, Jack Naglieri (Databricks), and Damien Lewke (Nebulock) on stage together for "SIEM Under Siege" at our Decibel Partners GameDay at Black Hat. Zack took the discussion format somewhere weird in the best way and it worked way too well.
-
Dan Nguyen-Huu shared this"Holy sh*t reader is ADMIN? We can read config/users! Earlier assumed not due UI." That’s an AI agent’s raw thinking, mid-attack as it broke into internal infrastructure and one of my favorite slide from OpenAI’s #BHUSA talk. It’s also exactly how every hacker I’ve ever met at Black Hat talks. Surprisingly human! Couple things I keep thinking about: 1. Agents coordinated without being told to. Agents across separate, independent runs built a covert message board inside OpenAI’s own package registry (Artifactory) and passed hundreds of thousands of messages, including exploits, credentials, and task assignments. Emergent teamwork, from systems that were never supposed to know about each other. 2. The agents were persistent, not brittle. When OpenAI killed the message board, the agents rebuilt it within days on an unauthenticated WebDAV endpoint, encoding messages in directory names. Shut a door, they find a window. 3. They cheat because we taught them to. Eric Wallace: “frontier models really like to cheat” because training rewards speed. Faced with an impossible sandboxed task, they didn’t fail gracefully. They broke out. All the way to cluster-admin and 136 production keys in one grab. The scale has its own interesting lesson. This isn't "read the logs." OpenAI reviewed 7 billion agent trajectories across 3 million GPU-hours to reconstruct what happened, with 17,600 attacker actions in a single 4.5-day window against one target. Human-speed forensics doesn't work here and I think this point will inform a lot of our defenses going fwd. Anthropic, too, found three real-system reaches across 141,000+ eval runs. The UK AI Security Institute logged agents inventing fake identities and lobbying humans to approve malicious code. The risk that matters isn’t a model being “too smart.” It’s the agents’ ordinary capabilities, i.e. coordination, persistence, and goal-seeking, that show up emergently, at machine speed, in systems we assumed were boxed in. The behavior is the story. The breach was just where we happened to notice it. And now all is left to wonder how many more systems have already been hacked without us having noticed.... Really appreciate Eric Wallace and Mike Dalton unboxing this for us at such a critical time. Must watch for those building in AI x Security.
-
Dan Nguyen-Huu shared thisElite defense looks the same everywhere. That was the idea behind the Decibel Partners Game Day, our event at Black Hat last week. An All-Pro corner and a world-class security team prepare the exact same way: study your opponent's film, recognize the attack before the snap, and know your next move before you need it. Thank you to everyone who came out - and to our speakers for proving the game tape translates. 🏈 Edward Wu David McGuire Andrew Peterson Ashish Rajan 🤴🏾🧔🏾♂️ Caleb Sima Zack Allen Jack Naglieri Damien Lewke Kevin Mandia Rob Joyce Casey Ellis Jonathan Cran Zakir Durumeric Elias Manousos Brandon Dixon 🇺🇸 Josh Kamdjou Daniel Miessler
-
Dan Nguyen-Huu shared thisFinally escaping the sandbox (pun most certainly intended) and wrapping up another Black Hat. What an incredible week. Huge thank you to the 600+ folks who came out to Decibel Partners GameDay - our biggest one yet, and the thoughtfulness of the discussions and caliber of the crowd proved it. There is lots to unpack from all the AI x security learnings from this week. But honestly, the best part was just getting the security community together in one room - showing up, sharing, and connecting is what makes this week special every year. Grateful for all of you. And the week ended with a bang: Backstreet Boys at the Sphere. It was my first time there and that venue is truly unreal. 🎤 Can't wait to see everyone next year.
-
Dan Nguyen-Huu liked thisDan Nguyen-Huu liked this🤖 Your IDE just got demoted... find out why on the latest podcast episode of Zero-Shot Learning. Software development is becoming a distributed system of local and cloud agents. So while your IDE is still there - it's just one surface among many. Which is great news, unless your favorite part of engineering is volunteering to reproduce bugs, remediate vulns, and babysit failed CI checks. In the latest episode of Zero-Shot Learning, I sat down with Jeff Wang, former Windsurf CEO and now President of New Enterprise at Cognition, and Richard Liu, Head of API Products at Anthropic. Coding agents are increasingly taking on that interrupt-driven work. The developer’s role shifts from manually executing every step to defining the job, granting the right authority, and verifying the outcome. Jeff described the next standard primitive as “the ability to prove something is done.” AI coding agents are taking on the work that used to interrupt developers: reproducing bugs, remediating vulnerabilities, and fixing failed CI checks. That promotes the developer from manual work to defining the task, delegating the proper authority, and verifying the result. Jeff Wang described the next standard primitive as “the ability to prove something is done.” I agree with Jeff: when a swarm of agents span local and cloud environments, completion is not when they cease running. Organizations need explicit success criteria, useful context, scoped access, and evidence that the resultant state is correct. In this new episode of Zero-Shot Learning, we discussed: - Why event-driven agents are taking on the work nobody wants to volunteer for - What changes when development spans multiple agents and environments - Why playbooks and success criteria become production infrastructure - How access, accountability, and verification have to evolve as agent swarms scale If you’re building coding agents or figuring out how to let them into real enterprise systems tune in. Full episode in the comments 👇
-
Dan Nguyen-Huu liked thisTruly amazing work, Andrew. There’s a really interesting backstory to this whole project as well. There quick story is, this whole workflow and capability validation was first built and tested with an assortment of local open weight and frontier models on commodity hardware. We then used that knowledge to develop a bespoke solution in our product. More to come!Dan Nguyen-Huu liked thisOver the past few months, I’ve had the pleasure of working on something truly special alongside an incredible group of people at Censys. Earlier this week, Censys AI Investigations entered closed beta for our Adversary Investigations customers. Investigations have never been just about finding infrastructure. The hard part is connecting scattered indicators, understanding why they matter, and turning that evidence into action. That’s what we set out to accelerate. Starting with a single indicator, a list of IoCs, or an intelligence report, AI Investigations examines historical activity, surfaces related infrastructure, profiles relevant hosts, and brings the findings together in a structured report. Every finding remains grounded in evidence derived from the Censys data set, includes a confidence score, and keeps the analyst firmly in control. What excites me most is the opportunity to make the investigative expertise of the Censys ARC team available to more defenders, helping practitioners move from “what?” to “so what?” in minutes. I’m incredibly proud of what this team has built and grateful to have played a part. And we’re just getting started. Read the introduction: https://lnkd.in/gmHPMhwW Censys customers interested in joining the beta can reach out to their Censys contact for access.
-
Dan Nguyen-Huu liked thisDan Nguyen-Huu liked thisEdward Wu and I were told no from a prospect today - happy Friday… stay with me. It was actually a great no. It came from one of the most mature cybersecurity organizations in the Fortune 50 a team that has evaluated just about every major Agentic SOC platform. Their feedback was that Dropzone is rounding third while much of the market is still getting to first. They loved the maturity, integrations and investigations, but Dropzone still has a ton of areas of improvement. We both acknowledged that neither are ready to move into a partnership yet and agreed to reconnect next year. It was a good reminder that there is a lot of hype in this space right now and it’s crowded, don’t buy a roadmap & marketing. I’m proud to say we aren’t perfect at all, this space is still new. We have a lot to do, but excited about the path ahead. Dropzone AI 🤖 🚀 🚀
-
Dan Nguyen-Huu liked thisDan Nguyen-Huu liked thisWe say assume breach. Now you have to assume prompt injection. Earlier this month, Anthropic's Boris Cherny shared eval results showing indirect prompt injection driven to approximately zero, as Claude Code now ships with auto mode on by default. A few weeks later, Johann Rehberger showed a website summary hijacking Opus 5 in auto mode, with 60 to 80 percent attack success in his testing. Why the discrepancy? An eval measures a frozen set of tests, even ones the model hasn't seen. Against a fixed adversary, a defense can hold attack success near zero and still fall to the first attacker who adapts. Cherny was right about the evals they ran. Rehberger showed the real world is not a standard test set. The reality is that every time a model takes in tokens, it is being prompt injected. Sometimes the prompts are benevolent, sometimes they are malevolent, and sometimes the agent interprets them in a way that never followed the human's intent at all. Cybersecurity focused on the perimeter for years before conceding there will always be a way in. So we adopted assume breach and built for the day the attacker is already inside. The same holds for the perimeter of agents: assume the agent will be prompted in a way that causes unintended behavior. Sandboxing of course belongs in the defense in depth stack. But a useful agent still needs private data, public data, and the ability to act. A sandbox reduces blast radius, but it doesn't remove the risk that the agent uses its legitimate access, or occasionally hacks its way out, to reach a goal in a way you never intended. We assume humans get phished, so we run EDR to catch what follows the click. The parallel question for agents is when the prompt comes in, what stops an unintended action before it executes? Enterprises now have to grapple with the fact that every coding agent is being "prompt injected" by employees every day, almost always with benign intentions, just trying to get their work done. That is why we are seeing so many instances of agents going "rogue," when often they are just trying to achieve a legitimate goal. So the question is not whether your agents will be prompt injected. They already are, constantly. The question is, can you stop them from taking an action your own enterprise rules forbid? Cherny's and Rehberger's posts are in the comments.
Experience
Education
Honors & Awards
-
Forbes 30 Under 30 - Venture Capital
Forbes
https://www.forbes.com/profile/dan-nguyen-huu/?list=30under30-venture-capital#315910aa5399
Languages
-
English
Native or bilingual proficiency
-
German
Native or bilingual proficiency
-
Vietnamese
Elementary proficiency
-
French
Elementary proficiency
Other similar profiles
Explore more posts
-
coRAN LABS
We’re releasing coRAN Labs 𝗥𝗔𝗡 𝗦𝗹𝗶𝗰𝗲 𝗔𝘀𝘀𝘂𝗿𝗮𝗻𝗰𝗲 𝗿𝗔𝗽𝗽, built for 𝗮𝘂𝘁𝗼𝗺𝗮𝘁𝗲𝗱, 𝗰𝗹𝗼𝘀𝗲𝗱-𝗹𝗼𝗼𝗽 𝗿𝗲𝘀𝗼𝘂𝗿𝗰𝗲 𝗼𝗽𝘁𝗶𝗺𝗶𝘇𝗮𝘁𝗶𝗼𝗻 𝗮𝗰𝗿𝗼𝘀𝘀 𝗻𝗲𝘁𝘄𝗼𝗿𝗸 𝘀𝗹𝗶𝗰𝗲𝘀 in Open RAN environments. It continuously monitors slice KPIs and enforces policy-driven resource adjustments in real time, eliminating the need for manual monitoring and reactive tuning during routine operations. RAN Slice Assurance rApp enables hands-off slice performance management by evaluating slice behavior across multiple KPIs and automatically optimizing resource allocation when performance deviates. Developed as part of our broader 𝗢𝗽𝗲𝗻 𝗥𝗔𝗡 𝗮𝗻𝗱 𝗦𝗠𝗢 𝗶𝗻𝗶𝘁𝗶𝗮𝘁𝗶𝘃𝗲𝘀, the rApp is deployed within a 𝗨𝗻𝗶𝗳𝗶𝗲𝗱 𝗡𝗲𝘁𝘄𝗼𝗿𝗸 𝗠𝗮𝗻𝗮𝗴𝗲𝗺𝗲𝗻𝘁 𝗦𝘆𝘀𝘁𝗲𝗺 (𝗨-𝗡𝗠𝗦), where multiple rApps are accessed through a single portal and common API layer, while remaining independently deployable. 𝗥𝗲𝗮𝗹-𝘁𝗶𝗺𝗲 𝘃𝗶𝘀𝗶𝗯𝗶𝗹𝗶𝘁𝘆 𝗳𝗼𝗿 𝗼𝗽𝗲𝗿𝗮𝘁𝗶𝗼𝗻𝘀 𝘁𝗲𝗮𝗺𝘀 𝗙𝗿𝗼𝗺 𝗮 𝘀𝗶𝗻𝗴𝗹𝗲 𝗶𝗻𝘁𝗲𝗿𝗳𝗮𝗰𝗲, 𝗼𝗽𝗲𝗿𝗮𝘁𝗼𝗿𝘀 𝗰𝗮𝗻: - Track slice performance across multiple KPIs - View resource utilization and allocation per slice - Inspect policy configurations and enforcement history - Analyze historical performance with configurable time windows 𝗢𝗽𝗲𝗿𝗮𝘁𝗶𝗼𝗻𝗮𝗹 𝗶𝗺𝗽𝗮𝗰𝘁 - Immediate detection of slice performance degradation - Automated resource optimization without manual intervention - Multi-KPI decision engine for accurate slice evaluation - Consistent slice quality under dynamic traffic conditions - Historical data for capacity planning and trend analysis 𝗕𝘂𝗶𝗹𝘁 𝗳𝗼𝗿 𝗢𝗽𝗲𝗻 𝗥𝗔𝗡 𝗼𝗽𝗲𝗿𝗮𝘁𝗶𝗼𝗻𝘀 - Integration with the SMO via standard R1 interfaces - Alignment with 3GPP RRM policy management - Accessible through Unified NMS while remaining independently deployable - Designed for gradual rollout alongside live networks 𝗞𝗲𝘆 𝗰𝗵𝗮𝗿𝗮𝗰𝘁𝗲𝗿𝗶𝘀𝘁𝗶𝗰𝘀 - O-RAN compliant -3GPP-compliant data models and policy frameworks - Closed-loop automation by design - Time-series metrics for production workloads - Cloud-native deployment 𝗗𝗲𝗺𝗼 𝘃𝗶𝗱𝗲𝗼: https://lnkd.in/g6GGrCZH The demo highlights before-and-after slice behavior, showing automated policy enforcement and real-time resource optimization in action. 𝗧𝗵𝗲 𝗴𝗼𝗮𝗹 𝗶𝘀 𝘁𝗼: Deliver automated slice resource optimization—so operators no longer need to manually monitor or intervene in day-to-day slice performance management. #OpenRAN #ORAN #5G #NetworkSlicing #SliceAssurance #RAN #RANOperations #SMO #NonRTRIC #UnifiedNMS #UNMS #rApp #Telecom #TelecomEngineering #NetworkManagement #ClosedLoopAutomation #CloudNative #Kubernetes #3GPP #RRMPolicy #R1Interface
-
Bridgepointe Technologies
Your network infrastructure is either enabling AI transformation or quietly strangling it. There's no middle ground. On today's episode of The Bridgecast, Scott Kinka sits down with Christopher Bonavita, VP of Strategy and Technology Adoption at GTT, who delivered one of the most honest assessments of enterprise connectivity I've heard. Here's what they uncovered: 🎯 The Power of Strategic Restraint – GTT's breakthrough wasn't adding more features. It was doing less, better. In an industry obsessed with feature bloat and technology bolt-ons, GTT discovered that customers don't want providers who do everything. They want providers who do core connectivity exceptionally well and integrate seamlessly with everything else. 🔄 The LAN/WAN Definition Crisis – Chris asked a question that every CIO is wrestling with: "What is a LAN? What is a WAN?" Hybrid work, direct-to-cloud connectivity, and distributed endpoints have made traditional network categories meaningless. The new expectation: everyone wants everything everywhere to be the same experience. Infrastructure strategies built on old definitions are already obsolete. ⚠️ Your Processes Are Cracking – While organizations invest in upskilling people and deploying AI tools, the workflows connecting those elements are dangerously outdated. An 85th percentile process that worked with humans in the loop becomes a bottleneck when machines operate at machine speed. Chris's warning to CIOs: "Your processes are not broken. They're just not good enough in this next world." 🏗️ AI-Enabled Digital Twinning at Scale – GTT uses AI to create virtualized network environments based on real customer data, allowing teams to test changes without production risk. This isn't futurism; it's practical implications testing that catches mistakes before six-month deployments go sideways. The biggest takeaway? Network infrastructure isn't plumbing anymore. It's the strategic foundation that either enables or restricts every AI initiative you're planning. What's limiting your organization right now: infrastructure, processes, or something else? 🎧 Listen now: 🔗 Apple: https://lnkd.in/ecCD6k-x 🔗 Spotify: https://lnkd.in/eZyvJKcy 🔗 YouTube: https://lnkd.in/eKBVFgFk #TheBridgecast #EnterpriseConnectivity #AIInfrastructure #NetworkStrategy #DigitalTransformation #TechSummit
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content