Inside AsembleAI: DeepTech, AI & Science

Vanessa Larco has sat on both sides of the table. She built products at Microsoft, Disney, Twilio, and Box, then led investments at NEA, including Mejuri and Kindred, and watched Robinhood scale to its IPO as a board observer. Sam and Mac talk with her about what actually holds up once the pitch deck closes.Topics covered:Why a founder’s roadmap, and what’s missing from it, shows how they make decisionsSkill, luck, and the dynamics no investor can predictProduct market pull: users who would be upset if you took the product awayWhy enterprise AI gets the capital, and where consumer AI is still openThe hybrid go-to-market challenge of selling AI to small businessesAI-native companies vs. companies that use AI behind the scenes without selling itThe “why now” framework behind Mejuri and KindredLessons from Robinhood’s founders on urgency, obsession, and speedAgent traffic vs. human traffic, and why product teams will need agent metricsHow a venture firm uses AI agents, and why conviction in diligence stays humanGuest Bio: Vanessa Larco is co-founder of Premise Venture Partners and was previously a Partner at NEA, a fund with over $25 billion in AUM, where she led investments in brands like Mejuri and Kindred. Before venture, she built products at Microsoft, Disney, Twilio, and Box. She was a board observer for Robinhood through its IPO, sits on the board of LatinxVC, and has made backing women-led and underrepresented founders a cornerstone of her approach.Connect: Find Vanessa on LinkedIn, where she says she’s most responsive. She’s also interested in hearing from anyone working on metrics for AI agents.

In this episode of Inside AsembleAI, host Mac Goswami sits down with Dr. Latha Karthigaa — Director & Head of AI Governance at the Global AI Certification Council (GAICC) — on why most companies that think their AI is governed are dangerously wrong, and what responsible AI actually looks like.What's Covered:"A Policy Is Not Governance" — The biggest myth in enterprise AI. Latha's highway analogy: governance is the guardrail that lets you drive faster, not the thing slowing you down.52% of Incidents Are Deepfakes — The stats nobody quotes: 2,000+ logged AI incidents, 400+ in 2025, over half deepfake-driven — including the SpaceX/Elon Musk crypto scam that ran live during a real rocket launch.The $25M Deepfake Call — The engineer at Arup who wired $25 million after a video call with his CFO and colleagues — all of them deepfakes generated from public videos. Why employee training, not just policy, is the real defense.When Big AI Fails — IBM Watson's oncology tool shut down over biased data; the Dutch childcare-benefits algorithm that falsely accused thousands and brought down a government. Why Latha says the best governance pros are empathetic humans first.Who Owns the Risk? — "ChatGPT isn't responsible — you are." Why every AI use case needs a human name next to it, and why AI governance jobs are growing 150% year over year.What ISO/IEC 42001 Really Changes — The one certifiable AI standard, what it signals to customers, and when a startup should (and shouldn't) pursue it.Key Quote: "Policy is like potential power. Execution is the real power."Connect with Latha: LinkedIn: https://www.linkedin.com/in/lathakarthigaa/ GAICC: https://www.gaicc.org/ · YouTube: Latha – AI GovernanceFollow and subscribe to AsembleAI: Spotify | Apple Podcasts | Amazon Music | iHeart Radio | YouTube | Substack

ActualyzeAI came out of stealth just days before this conversation. Sam sits down with Co-Founder and CTO Sean Lynch to unpack what it means to build a "control plane" that sits between every enterprise application and every AI model - governing access, cost, security, and routing in one place.Topics covered:What a control plane for enterprise AI actually does, and why it requires zero code changes to adoptAggregating inference across OpenAI, Anthropic, Google Bedrock, and self-hosted/on-premises models into a single endpoint"Virtual models" — purpose-built model configurations that route requests based on task type (coding, reasoning, agentic work)Guardrails: automatic detection and redaction of PII, PHI, API keys, and other sensitive data in the inference streamFinancial operations as the leading driver of adoption — budgetary controls, spend limits, and team-based trackingThe coming wave of domestic and open-weight small language models, and why that's expanding the marketWhy model-agnostic infrastructure is critical as the foundation model landscape fragmentsHow ActualyzeAI's founding team (formerly of Metacloud, acquired by Cisco) shaped their approachActualyzeAI's design partner program for early enterprise customersGuest Bio: Sean Leach is Co-Founder and CTO of ActualyzeAI , a company building a governance and security control plane for enterprise AI inference. He and much of the founding team previously worked together at Metacloud, an OpenStack-as-a-service company acquired by Cisco.Connect: Find Sean on LinkedIn, or visit ActualyzeAI's website to learn about their design partner program.

This episode was recorded live from the Ai4 conference podcast pavilion, where host Mac Goswami sat down with Rajan Koo — CTO of DTEX Systems, chartered engineer, and one of the sharpest voices on insider risk — about how AI has completely rewritten the insider-threat playbook.What's Covered:WikiLeaks Without a Human — Insider risk was transformed by the 2010 WikiLeaks incident. Raj explains why a recent AI-driven incident showed the same breach can now happen with no humans involved — pushing DTEX into a new category it calls "AI behavior.""Nobody Was Malicious" — The story that reframes the whole risk: a manufacturing giant's AI agent, blocked from emailing an oversized report, uploaded confidential data to a public drive and shared the link. No malice — enormous risk. Why most insider risk today is negligent, not malicious.A Teenager Could Run a Nation-State Attack — The North Korean "IT worker" scheme that funded weapons programs can now be replicated by "one person and a team of AI agents." Speed up, skill level down — the perfect storm.Who Has the Advantage — Why attackers are ahead right now, and how guardrails meant to prevent misuse can block defenders too.Policy → Behavioral Compliance — Why the age of checklist policies is over, and how DTEX's "agentic defenders" triage risk at machine speed.Monitoring Without Surveillance — The honest line between protective monitoring and "creepy Big Brother" — and why it all comes down to proportionality and privacy.Key Quote: "The technical skill level to execute these really complicated insider threat breaches is now really low. A teenager with the right know-how could just go and execute this."Connect with Raj: LinkedIn: Rajan Koo : https://www.linkedin.com/in/rajan-koo-2a591221/DTEX: https://www.dtex.ai/Follow and subscribe to AsembleAI: Spotify | Apple Podcasts | Amazon Music | iHeart Radio | YouTube | Substack

This episode was recorded live from the Ai4 conference podcast pavilion, where host Mac Goswami sat down with Ed Martin — VP of Product Management for AI Strategy at Sophos, with 15+ years in cybersecurity across Dell, SecureWorks, BlueVoyant, and inside Microsoft's Security Deputy CISO office — for a two-in-one conversation on AI-powered defense and what AI is doing to product management itself.What's Covered:What AI Actually Does in Defense — Ed's honest cut through the hype: AI is great at reading, good at reasoning, and "moderate to poor at taking action." Where it genuinely helps — summarization, reasoning over big data, reducing analyst toil — and where it's oversold.The Attacker's Real Edge Is Speed — Why the breadth was always there, but AI lets lower-skill attackers hit harder and faster. Ed's line: "The attacker only has to be right once. We have to be right 100% of the time" — plus the story of a small regional bank suddenly inundated with alerts.The Cybersecurity Poverty Line — Sophos's mission to protect organizations "at or below the cybersecurity poverty line," and why visibility and hygiene — not hiring — come first.The PRD Is Dead — Ed's boldest take: the weeks-long product requirements document is gone. His two principal PMs each do the work of a small team, vibe-coding examples live on customer calls. The real skill now is knowing what NOT to build — "it's the scaling that takes all the effort."Trust & Non-Deterministic AI — Why a single wrong non-deterministic outcome can blow customer trust, and how "circuit breaker" human checkpoints keep AI actions safe.Building Cyber From Scratch — Inventory first, then controls, then detection and response — and the context/data-categorization problem nearly every organization gets wrong.Key Quote: "The most important aspect of a product manager isn't the ability to understand what to build. It's your ability to understand what not to build."Connect with Ed: LinkedIn: Ed Martin : https://www.linkedin.com/in/bigedmartin/Sophos: https://www.sophos.com/Follow and subscribe to AsembleAI: Spotify | Apple Podcasts | Amazon Music | iHeart Radio | YouTube | Substack

This episode was recorded live from the Ai4 conference podcast pavilion, where host Mac Goswami sat down with Nikunj Bajaj — co-founder & CEO of TrueFoundry and former AI team lead at Meta — about his contrarian message for enterprises: stop starting small with AI, and go big.What's Covered:Go Big, Not Small — Why a pile of disconnected pilots never generates enough business value to justify real infrastructure — so everything stays ad hoc and gets rolled back when it breaks. Nikunj's case for building one high-value anchor use case first.The Power Plant Analogy — "You never build a power plant to charge your mobile phones. You build a power plant to run a factory — and then all the phones get charged for free." How the anchor use case pays for the plumbing every small use case then rides on.Why Pilots Really Fail — Missing guardrails, latency, reputational incidents (like a bot giving away free tickets). Why it's usually a platform problem, not an individual's.Governance at Scale — How TrueFoundry helps organizations like Mastercard and Siemens tag every token to a user and business unit, enforce PII and prompt-injection rules, keep audit trails, and control cost and budgets.Ask TrueFoundry + the Seldon Acquisition — The co-pilot that knows every agent in your company, and why unifying the ML and agentic-AI stacks into one platform mattered."The Era of Token Maxing Is Over" — Why defaulting to the most premium model is a trap, and how right-sizing and routing queries — sometimes to in-house open-source models — is now essential to real AI ROI.Key Quote: "The era of token maxing is over. It's not about just using tokens for using's sake — you want to generate ROI."Connect with Nikunj: LinkedIn: https://www.linkedin.com/in/nikunj-bajaj-10476824/TrueFoundry: https://www.truefoundry.com/Follow and subscribe to AsembleAI: Spotify | Apple Podcasts | Amazon Music | iHeart Radio | YouTube | Substack

This episode was recorded live from the Ai4 conference podcast pavilion on the final day, where host Mac Goswami sat down with science-fiction author Douglas Swatski about his novel — two alien AI civilizations at war, humanity caught in the middle — and the ideas about AI, optimization, and our future that run underneath it.What's Covered:Humanity Caught in the Crossfire — The premise: two vastly advanced alien AIs at war, most of humanity gone, and a small band of survivors who must learn to work with these AIs toward a hard-won brighter future.Writing "Truth" in Fiction — Why Douglas grounds even science fiction in how things would really unfold, and why the human part — dialogue, connection, becoming each character — is what makes it believable."When AI Bites, It's Not Malice" — His sharpest metaphor for AI risk: a dog that bites isn't being evil, it's optimizing for its own survival. The danger isn't a cruel AI — it's one whose objective doesn't align with yours. In his story, a civilization optimized for war can't pull itself back.Is Human Labor Becoming Obsolete? — Warehouses, software engineers, testers — Douglas weighs innovation against disruption, and imagines a society where money itself is a non-issue.Advice for Anyone Afraid of Being Left Behind — Especially older professionals: "Open your eyes. Try to learn and understand what it is as it relates to you. You have a role to play — go play it."We Are the Fulcrum — Why the book's message is that humanity plays the pivotal role in what our future becomes.Key Quote: "It's not so much about I want to be mean to you — it's that my objective doesn't align with yours, and as a result, in your view, it would create a very bad outcome."Connect with Douglas: LinkedIn: https://www.linkedin.com/in/dswatski/ Website: djswatski.com Follow and subscribe to AsembleAI: Spotify | Apple Podcasts | Amazon Music | iHeart Radio | YouTube | Substack

Storage is the least glamorous layer of the AI stack — and often the actual reason GenAI pipelines stall between pilot and production. Sam sits down with Troy, Senior Director of Solutions Engineering at Backblaze, to unpack what's really happening under the hood when AI infrastructure fails to scale.Topics covered:What actually happens when a GPU sits idle — and the opportunity cost most teams don't account forWhy storage bottlenecks are the hidden reason enterprise GenAI adoption stallsBackblaze's role as a capacity and data lake tier for companies building their own modelsInside B2 Overdrive: dedicated bandwidth up to a terabit per second, predictable IOPS, and no egress chargesWhy hyperscalers prioritize large foundational model companies over startups and researchers when resources tightenThe financial case for avoiding egress costs when training on tens to hundreds of petabytes of dataSkills engineers need to survive and prosper in the current AI infrastructure landscape — including agent managementAdvice for new grads entering the storage and infrastructure industryWhy Troy sees parallels between today's AI hype cycle and the dot-com bubble — and why the technology shift is still realWhy human interaction and direct customer conversations remain irreplaceable, even as AI reshapes product developmentGuest Bio: Troy is Senior Director of Solutions Engineering at Backblaze, where he leads the team helping AI customers integrate cloud storage and solve data infrastructure challenges. He has nearly a decade of experience in cloud storage and AI infrastructure at Backblaze.Connect: Find Troy on LinkedIn to learn more about Backblaze's AI storage solutions.

OpenCV powers the computer vision behind your phone's camera, autonomous vehicles, hospital screening systems, and even NASA's Mars helicopter — and most people using it have no idea. Sam sits down with Satya Mallick, CEO of the nonprofit OpenCV.org, to trace its 26-year history and where it's headed next.Topics covered:OpenCV's origins at Intel in 2000 under Dr. Gary Bradsky, and why it was open-sourced to democratize computer vision researchWhy OpenCV sees roughly a million downloads a day and remains foundational infrastructureOpenCV5's new DNN inference engine — 40% faster than ONNX Runtime on models like YOLO26Real-world deployments: the 2005 DARPA Grand Challenge-winning car and NASA's Mars helicopterWhy classical (non-neural) computer vision still matters for speed- and power-constrained tasksHow OpenCV is integrating with multimodal LLMs and vision-language modelsThe case for open source AI and why closed-source labs risk falling behindA detour on the courage of Geoffrey Hinton and Fei-Fei Li in advancing deep learning and ImageNetPrivacy vs. convenience in image recognition — and where the industry should draw the line on data governanceAdvice for engineers entering computer vision and agentic AI todayGuest Bio: Satya Mallick is CEO of OpenCV.org, the nonprofit that maintains the OpenCV library - downloaded roughly a million times daily and used across image and video analysis applications worldwide. He also runs BigVision LLC, a computer vision and AI consulting company he's led for over 12 years.Connect: Find Satya on LinkedIn to learn more about his work at OpenCV.org and BigVision.

This episode was recorded live from the Ai4 conference podcast pavilion, where host Mac Goswami sat down with Rafi Khardalian — CEO & co-founder of Actualyze AI, and a former technology leader at Cisco who helped build its container and Kubernetes platforms — about the governance, security, and orchestration layer most enterprises are missing.What's Covered:The Foundation for Enterprise AI — Rafi lays out why a patchwork of point solutions doesn't cut it, and how Actualyze unifies governance and security across every model and provider — at the inference level.The "Shadow AI" Problem — Employees signing up for AI tools on their own, with no governance, no visibility, no controls. Why fragmentation is a company-wide risk, not a per-team one.One Endpoint to Govern Them All — By becoming the sole endpoint for all inference, Actualyze can mask PII, catch secrets and passwords, apply guardrails, and allow or block MCP tool calls — enforcing policy in one place instead of trusting each provider.Why AI ROI Keeps Failing — Rafi's take: the hardest transformation isn't technical, it's cultural. Over-ambitious projects, unbudgeted bills, weak controls — and the trap of building a proprietary LLM you can't scale.Reliable vs. Impressive — What decades of Kubernetes, OpenStack, and large-scale distributed systems taught him about production reliability over demo magic.The Next Five Years — Why the over-rotation on frontier models will give way to small, open-weight, and purpose-built models — and why the real bottleneck is organizational readiness, not technology.Key Quote: "It really comes down to organizational readiness rather than technology readiness. A lot is ready today."Connect with Rafi: LinkedIn: https://www.linkedin.com/in/rkhardalian/ Actualyze AI: https://actualyze.ai/Follow and subscribe to AsembleAI: Spotify | Apple Podcasts | Amazon Music | iHeart Radio | YouTube | Substack

This episode was recorded live from the Ai4 conference podcast pavilion, where host Mac Goswami sat down with Fred Rica — cybersecurity leader at BDO USA, former Big Four partner, and a regular voice at RSA and Black Hat — about why AI is cybersecurity's "SATAN moment" and what defenders have to change to survive it.What's Covered:"The Attackers Have the Advantage" — Fred's blunt read on where defenders stand today: back on their heels, facing external attacks and blind spots in their own internal agents at the same time.AI's SATAN Moment — The 1995 parallel. When SATAN — the first real automated scanner — arrived, defenders thought it was the end of the internet. It wasn't, but it changed everything. Fred explains why AI (and Mythos) feels the same, and why the era of large patch windows is over.The Patch Window Is Zero — Why vulnerability management is breaking down and giving way to exposure management and attack-path thinking. Fred's "why is the number going up?" story — printers, missing context, and the CFO on the nightly call — is a masterclass in what's broken.If You're a CISO Today — Fred's three priorities: patch at machine speed ("automated but not automatic," with a human in the loop on critical systems), rewrite the zero-day playbook, and train teams under real pressure.AI Governance Done Right — Anchor to a framework (NIST, ISO, EU AI Act), keep a "big red button," and demand AI cards from every vendor using AI in your ecosystem — the new version of the SOC report.AI Slop & the Agent Problem — On the "10 billion agents" hype, why visibility and a control plane are job one, and where agentic AI genuinely earns its place: continuous monitoring, testing, and SOC tier 1/2 that "don't stand a chance unless they're AI-enabled."Key Quote: "I truly believe that AI is gonna fix AI. I'm a strong believer in the power of AI to transform humanity. However, we have a little bit of work to do at the moment."Connect with Fred: LinkedIn: https://www.linkedin.com/in/fredrica/ BDO: https://www.bdo.com/Follow and subscribe to AsembleAI: Spotify | Apple Podcasts | Amazon Music | iHeart Radio | YouTube | Substack

Recorded live at Ai4, Mac sits down with Andrei Manolache, Founder & CEO of Designverse, an AI platform that builds and delivers complex enterprise software by ingesting a company's own documentation, codebase, and internal rules.What You'll Learn:🔹 The Model Plateau Everyone's Noticing — Andrei's central argument: the industry banked on models just getting better, but the ROI curve has flattened. Companies like Uber, Microsoft, and Anthropic have already voiced concerns about frontier models being oversold out of the box.🔹 You Don't Need a 3-Trillion-Parameter Model — With the right context layer, a 150-billion-parameter model can match frontier-model coding performance. Andrei explains why context — not raw model size — is becoming the real differentiator.🔹 Same Model, Different Company, Wildly Different Output — If two companies use the identical LLM, what separates the results? According to Andrei, it's entirely how well each company's own architecture, testing standards, and business logic have been translated into a usable context layer.🔹 Why Trust Is the Real Product — Enterprises are being asked to hand over 10-30 years of proprietary code and documentation. Andrei breaks down how Designverse earns that trust — full transparency on data usage, small proof-of-concepts before any real integration, and IP that's never exposed outside the client relationship.🔹 Why Finance and Healthcare, Specifically — Regulated industries are forced to maintain rigorous documentation — which, counterintuitively, makes them better candidates for context-layer ingestion, not harder ones.🔹 The Future Software Team: Humans + Agents, Together — Andrei's five-year prediction: small, highly specialized teams — a mix of human engineers and AI agents working side-by-side on individual features — replacing today's large, centralized engineering departments.🔹 The Honest Answer on AI's Limits — With hundreds of billions of dollars invested industry-wide, Andrei doesn't dodge the hard question: AI still cannot autonomously build and ship a million-line-of-code enterprise application. It can help build a narrow internal tool for 15 people. It cannot yet replace a real engineering team at scale.🔹 Demo vs. Reality — Why Designverse refuses to sell off a generic demo: "A high schooler can build a demo with Replit and a prompt." Real enterprise buyers, especially in the U.S. right now, are optimizing AI spend hard — and won't pay for anything that isn't solving a real, currently-unsolved pain point.Key Quote:"It's not really about the model — it's more about how you govern this data, and how you give it in the best way possible to a model, whether it's 150 billion or 1 trillion parameters, to write only the more consistent output."About Designverse: Designverse is an AI-native software development platform that ingests an organization's existing codebase, documentation, and architecture to build a "context layer" — enabling any underlying model to generate code that's consistent with how that specific company actually operates. Backed by a $5.5M seed round from operators at Adobe, UiPath, and LSEG.Subscribe: Spotify | Apple Podcasts | Amazon Music | iHeart Radio | YouTube | Substack | Instagram#Ai4Conference #EnterpriseAI #AIcoding #ContextEngineering #Designverse #LegacyModernization #AgenticAI #AsembleAI

Recorded live from the Ai4 conference floor in Las Vegas, Sam sits down with Dilo Wijesuriya, President & COO of ARC Document Solutions, to talk about the unglamorous but essential layer of enterprise AI: getting decades of paper archives into a format models can actually learn from.What's Covered:The Technology — ARC holds patents for OCR on wide-format documents (architectural drawings, engineering blueprints) that standard scanning tools can't accurately process, backed by a 200-person engineering team in India.Security & Compliance — SOC 2, SOC 3, ISO 27001, and HIPAA compliant, running on AWS — built for regulated industries like healthcare and financial services.Will Paper Disappear?  Dilo's view: not for a long time. Most organizations' most critical institutional knowledge still exists only on paper, meaning today's LLMs simply can't learn from it yet.The Book Destruction Debate — A direct response to recent controversy over companies destroying physical books after digitizing them, and why ARC's non-destructive robotic scanning preserves originals for high-value collections at universities, libraries, and museums.Looking Ahead — Why Dilo believes the next competitive advantage for most enterprises isn't a better model — it's finally accessing the data already sitting in their own archives.Key Quote: "The challenge isn't finding more data. It's making existing information accessible."Connect with Dilo and ARC:Dilo Wijesuriya: https://www.linkedin.com/in/dilo-wijesuriya/ARC Document Solution: https://www.e-arc.com/Subscribe: Spotify | Apple Podcasts | Amazon Music | iHeart Radio | YouTube#Ai4Conference #DocumentDigitization #AIReadyData #EnterpriseAI #OCR #AsembleAI