Dr. Niklas: Venture Grade

Niklas
  • 0.0 (0)
  • Business
  • Updated Biweekly

Welcome to Dr. Niklas: Venture Grade - the intersection of deep tech and high-stakes business. I’m Dr. Niklas, an engineer, founder, and researcher. I look at business and technology through the lens of a builder: how do you architect a company that is robust, scalable, and "Venture Grade"? On this channel, I move beyond surface-level startup advice. I sit down with world-class founders, investors, and technical leaders to debug their strategies and examine the source code of their success. What is "Venture Grade"? It’s a standard. Just like military-grade hardware or enterprise-grade software, a Venture Grade business is built to survive, scale, and secure investment. New episodes ship every Tuesday. Subscribe for weekly deep dives with people building the future 🚀☘️

  1. 1d ago

    DN #34: A Digital Brain, Privacy-First Memory & 1,000 Reddit Downloads (w/ Ahmet & Ömer Delibaş)

    "Your data has always been used for ads, for big data, for selling to other companies. And the user benefits the least. It doesn't have to be like that." I talk to Ahmet and Ömer Delibaş, two brothers building Klara, a "digital brain" for Android that captures everything you do on your phone, organizes it, and gives it back to you as searchable memory, habits, and a daily digest. One brother is a software engineer who spent years watching user data flow to ad networks and data centers. The other is a dentist who builds on the side. Together they spent four years turning an idea into a product. We get into the privacy model that most AI apps won't touch (fully local if you want, anonymized and deleted after five days if you turn AI on), how a chaotic accessibility feed becomes a clean memory database, the ADHD users who showed up unplanned, and the Reddit playbook that got them 1,000 pre-registrations before launch and 1,000 downloads in week one. In this episode: - Your Digital Brain: What Klara captures, what it throws away, and why your biological brain is the model. - Users Benefit the Least: The founding thesis, and why the AI wave made it buildable. - Five-Day Deletion: Local-only mode, anonymized AI calls, and leaving nothing behind when a user quits. - The ADHD Surprise: Users the founders never planned for, and "fall in love with the users, not the product." - 2K MAU, 10% MoM, Zero Ads: Why they refuse to buy growth before the product deserves it. - The Reddit Playbook: 1,000 pre-registrations in one month, 1,000 downloads in week one, and posting about more than Klara. - Great Product, Terrible Business: My Wunderjob lesson on churn, and why subscriptions beat ads for privacy products. - Focus on Hard Things: Why everyone builds AI agents, why memory is harder, and why distribution still beats product. 🎧 Full episode on all podcast platforms 💬 Would you let an app record everything on your phone if it deleted it after five days? Let us know in the comments! 🔔 Please like and subscribe! Every subscriber helps our channel grow. #DN #DigitalBrain #Klara #PrivacyFirst #AIMemory #AndroidApp #BuildInPublic #Reddit #IndieHackers #Startups Timestamps: 0:00 Intro 0:10 What Clara is: your digital brain 1:06 The engineer's path: years of watching user data go the wrong way 3:30 The dentist who builds: why two brothers started a company 4:41 Android first, other platforms later 5:17 The privacy model: local-only, anonymized, deleted in five days 9:28 How memory works: turning accessibility chaos into a database 13:14 Text only for now, and where AI devices are heading 13:51 The ADHD users nobody planned for 15:55 Building in public: X, Reddit, Telegram, in-app surveys 17:12 Wunderjob: a great product that was a terrible business 18:54 2K monthly active users, 10% growth, no ads yet 20:25 Subscription vs ads, and why big companies didn't start with ads 22:41 The Reddit playbook: 1,000 pre-registrations before launch 26:04 Advice for two-person teams: ship, distribute, focus on hard things

  2. Sep 22

    DN #33: Make the Internet Weird Again, The Flash Era & Change the Lovable Look (w/ Chus Margallo)

    "People will go to a website because they're going to enjoy the experience there, not because the information they need is there." I talk to Chus, a Barcelona-based art director and UX/UI designer with 15+ years across Madrid ad agencies, an Amsterdam mobile startup, a gaming corporate, a trading platform, and now a Dutch insurer, about his side mission to make the internet weird again. By day he designs AI tools for insurance agents at Nationale-Nederlanden. By night he builds hyper-realistic folder icons and a portfolio that runs like a 90s operating system. We get into why the Flash era was the best the web ever looked, how conversion obsession and design systems flattened everything, why he uses Lovable constantly but never ships the Lovable look, and his exact workflow from ChatGPT tech-stack advice to plan mode to telling the AI to stop adding emojis. In this episode: - Make the Internet Weird Again: Why AI answering your questions means websites now have to be worth visiting for the experience. - The Flash Era: What 2004 got right, and how accessibility and conversion obsession turned the web flat and minimal. - Bootstrap, Tailwind, Lovable: The same trap three times. Out-of-the-box defaults are why every AI site looks identical. - Never Ship the Lovable Look: Using AI for the code you can't write, then replacing every visual default with your own. - Agency vs Corporate: 15 years from late-night Madrid briefings to laser-focused insurance product teams, and why both can work. - AI in Insurance: How a regulated Dutch insurer studies China's AI moves and builds agent tools without breaking the rules. - The 140 Folders: From a post on X to a Lovable-built gallery and mini-shop, step by step. - Intentional Design: A Comic Sans can work if it's on purpose. Limit the AI early, upload your own fonts and icons. - Advice for Non-Designers: Learn the design movement behind what you call "nice" before you prompt anything. 🎧 Full episode on all podcast platforms 💬 Was the Flash-era internet actually better, or is it just nostalgia? Let us know in the comments! 🔔 Please like and subscribe! Every subscriber helps our channel grow. #DN #WebDesign #Lovable #UXDesign #MakeTheInternetWeirdAgain #ArtDirector #AIDesign #VibeCoding #DesignSystems #Startups Timestamps: 0:00 Intro 0:39 Madrid ad agencies: briefings, late nights, the golden rule 3:00 The jump to a mobile startup in Amsterdam 4:38 Learning how developers think without learning to code 6:00 Gaming, agencies, Barcelona, and the trading platform 7:00 Nacar design studio and the biggest corporate job yet 8:50 Make the internet weird again 9:55 Why the Flash era was the best the web ever looked 11:10 Nobody visits websites for information anymore 12:04 Design systems and why every website looks the same 13:31 Airbnb, Shopify, and the designers fighting back 14:45 Bootstrap, Lovable, and the out-of-the-box trap 16:12 Niklas's brokerage app: when a redesign kills the UX 18:58 Is this the NFT bubble again? 20:14 Local AI, open weights, and where the craft goes 22:43 Firefly, prompting, and why a designer now feels like a coder 26:35 AI inside a regulated insurer 29:21 The 140 folders: from an X post to a Lovable shop 32:50 Plan mode first, small prompts, stop the emojis 35:38 Advice for non-designers who want to build something nice

  3. Sep 8

    DN #32: Open Weight Models, Airbnb for Compute & Why Nobody Works Less With AI (w/ Alexander Solod)

    "I haven't met anybody who's working less after the release of AI. If anything, deadlines got shorter." I talk to Alexander Solod, an AI developer who has been building with generative models since a GPT-2-written paper blew his mind in college, about where AI actually stands right now: open weight models running at home, on-device chips, the interface problem nobody has solved, and what AI is really doing to jobs. We get into why Alex runs Gemma 4 on a Mac Studio for his personal agents, the "Airbnb for compute" idea of renting out your idle hardware for tokens, why radiologists are still booming ten years after AI was supposed to replace them, and the only career advice that survives automation: get into a position where you make decisions. In this episode: - Open Weights at Home: Why the future is small on-device models for daily work, with frontier models reserved for the hard stuff. - The Benchmark Problem: Why "better than the average human at using a computer" tells you almost nothing. - Filter and Enzyme: LLMs won't discover relativity, but they filter the world's information and collapse weeks of work into hours. - Airbnb for Compute: Renting out your idle Mac Studio for tokens, and why solar at home makes it obvious. - Chat Was V1: Voice in, visuals out, generative UI, and why the AI interface is still unsolved. - Ambient AI Scribes: The easiest healthcare win so far, and why radiology still hasn't been automated. - Nobody Works Less: The paradox of AI productivity, and why the layoffs blamed on AI are really pandemic overhiring. - The Job That's Safe: A machine can never be held responsible, so get into a position with stakes and decisions. 🎧 Full episode on all podcast platforms 💬 Are you working less since AI, or more? Let us know in the comments! 🔔 Please like and subscribe! Every subscriber helps our channel grow. #DN #OpenWeightModels #OnDeviceAI #AlexanderSolod #AIJobs #HealthcareAI #AIAgents #DecentralizedCompute #LocalLLM #Startups Timestamps: 0:00 Intro 0:25 The GPT-2 paper that pulled Alex into AI 1:12 OpenAI Gym and the hide-and-seek video 3:28 Open weight models: running Gemma 4 on a Mac Studio 4:04 Nobody knows the limit of intelligence 4:53 Is "better than the average human" a real benchmark? 6:49 Could an LLM have discovered relativity? 8:39 LLMs as filter and enzyme 9:54 On-device AI and Apple's neural engine bet 11:36 Airbnb for compute: renting out your idle hardware 13:35 Chat was V1: the unsolved AI interface 15:43 How far away are brain interfaces? 16:52 Predictive healthcare and the radiology myth 19:28 Giving clinicians their time back 20:48 Regulation, trust, and security in healthcare AI 22:24 Ambient AI scribes: the easiest win so far 23:47 Whoop data and democratizing Bryan Johnson 26:05 Nobody is working less since AI 27:23 Are the layoffs really AI, or pandemic overhiring? 30:31 How to prepare: actually use the agents 32:08 Best advice Alex ever received: keep momentum 33:01 Hot take: the AI interface is still wide open

  4. Aug 25

    DN #31: Agentic TPM, Temporal Knowledge Graphs & The Job AI Can't Replace (w/ Jake Kim)

    "People think TPM is gonna be the job that gets replaced first. In reality, it's gonna be the hardest thing to replace." I talk to Jake Kim, founder of Serro AI, about agentic technical program management and why the horizontal coordination role everyone expects AI to kill first is actually the one it can't. Jake was an engineering manager at Apple, where his team's software sat on the critical path to every iPhone and iPad launch, and where he watched world-class TPMs hold the whole machine together. We get into the temporal knowledge graphs that give AI agents organizational memory, the client with 2 TPMs, 300 people, and 75 recurring reports, the R&D tax documentation wedge that pays for the product on day one, and the founder lesson Jake learned right before running out of money. In this episode: - The Job AI Can't Replace: Why vertical agents (coding, design, product) make the horizontal coordinator more essential, not less. - Agentic TPM: Technical program management, not manager. Extending human TPMs' reach to the agents downstream. - Temporal Knowledge Graphs: Why programs need a memory of decisions, tradeoffs, and dependencies over time, not just a snapshot. - 2 TPMs, 300 People, 75 Reports: The middle-management squeeze nobody's tooling has touched. - The R&D Tax Wedge: The audit-ready documentation report worth millions in credits that gets Serro in the enterprise door. - Per-Program Pricing: Why Serro charges $100 per tracked program with unlimited seats, and what that alignment does in enterprise sales. - Don't Be Shy About Your Vision: The lesson Jake learned right before running out of money, and why humble framing nearly killed the company. - Advice for TPMs: Why the role should feel confident, not threatened, and how to extend your influence to the agent layer. 🎧 Full episode on all podcast platforms 💬 Is middle management really the hardest thing for AI to replace, or the first to go? Let us know in the comments! 🔔 Please like and subscribe! Every subscriber helps our channel grow. #DN #AgenticTPM #KnowledgeGraphs #JakeKim #SerroAI #AIAgents #EngineeringManagement #EnterpriseAI #Startups #TPM Timestamps: 0:00 Intro 0:15 What agentic TPM actually is 1:37 From Apple engineering manager to founder 3:12 Why AI employees haven't worked so far 4:46 Why agentic TPM is technically hard (programs vs projects) 8:22 Integrating with the company's tools 9:21 2 TPMs, 300 people, 75 recurring reports 13:15 Temporal knowledge graphs explained 14:28 How the product works: Gantt charts, drift alerts, MCP 16:00 Two years of pivots and what they taught him 17:50 The vision lesson: don't be shy right before running out of money 19:44 The R&D tax documentation wedge (and PwC) 22:38 Per-program pricing and the death-of-SaaS question 25:58 Why not just be a knowledge graph company? 27:54 The AI-native org two to three years

  5. Aug 18

    DN #30: Boring Industry Gold, AI Blueprint Readers & The Lead-Selling Trap (w/ Vinh Luu)

    "A good assistant takes about three days for one blueprint. I'm using AI, and it's 30 seconds." I talk to Vinh, solo founder of HD Parking Lot, about the construction marketplace he's building after a $25,000 junk-removal quote pushed him into the most overlooked trillion-dollar industry in tech. Vinh spent 20+ years as a software engineer before a rough patch (losing his mom, getting laid off) turned into a junk-removal side hustle, which turned into discovering how broken the contractor-homeowner relationship actually is. We get into why Angi and Thumbtack sell your information to 20 contractors and then abandon you, the AI blueprint reader that compresses three days of estimation work into 30 seconds, and why the boring industries nobody wants to touch are where the real opportunities hide. In this episode: - The $25K Junk Quote: The cleanup bill that turned a laid-off engineer into a junk removal operator, and then a construction founder. - The Lead-Selling Trap: Why Angi and Thumbtack make homeowners repeat themselves to 20 contractors, and what a marketplace that doesn't sell leads looks like. - The AI Blueprint Reader: Materials, labor hours, schedule, and regulations from an uploaded blueprint in 30 seconds instead of 3 days. - Boring Industry Gold: Why construction is a trillion-dollar industry tech people won't touch, and why that's exactly the opportunity. - The Full Lifecycle Bet: Directory, project bidding, materials marketplace, calculators, and community in one platform that stays past the handshake. - The Visualizer: The free tool that repaints your room from a photo in 30 seconds, built to settle an argument with his wife. - Don't Do Paper Research: Why one day living inside the process you want to automate beats a hundred customer interviews. - Kitesurfing as Focus Training: Why you can't ride the wind while worrying about your startup, and what that teaches about adaptability. 🎧 Full episode on all podcast platforms 💬 Would you rather build in the crowded AI market, or own a boring trillion-dollar industry? Let us know in the comments! 🔔 Please like and subscribe! Every subscriber helps our channel grow. #DN #ConstructionTech #AIBlueprints #Vinh #HDParkingLot #BoringIndustries #SoloFounder #BuildingInPublic #Startups #PropTech Timestamps (RAW — re-time after edit): 0:00 Intro 0:35 From junk removal side hustle to construction tech 4:14 The lead-selling trap: how Angi and Thumbtack actually work 8:33 The $25K junk quote that started everything 11:32 Why boring industries beat flashy AI startups 13:03 The AI blueprint reader: 3 days to 30 seconds 14:40 The Visualizer: repaint your room from a photo 18:37 The full feature walkthrough: directory, bidding, materials, calculators 26:26 Why he builds in public (the Marc Lou inspiration) 30:16 Marketing as a solo founder with an SEO background 33:32 Building a global company with an equity-based team 44:14 The one advice: hands-on research, not paper research 46:51 Kitesurfing, focus, and adaptability

  6. Aug 11

    DN #29: The Viral Loop of Imagination, Kids as Creators & Learning Through Story (w/ Feifei Qiu)

    "Kids don't just consume technology, they create with it." I talk to Feifei Qiu, founder of Kindlewood Studios, about the dragon story her four-year-old told at the dinner table that turned into a learning platform, and why storytelling is the wedge into teaching kids reading, writing, and creativity. Feifei spent years as a product manager at Microsoft before leaving to build Kindlewood for her own kids, and 200+ young storytellers later, the thesis is holding. We get into the viral loop of imagination (one kid's story inspires five more), how she designs AI that keeps kids in the driver's seat instead of writing for them, and the product lessons she carried over from enterprise PM work. In this episode: - The Viral Loop of Imagination: How one dragon story at a play date turned every kid in the room into a storyteller. - Kids as Creators: Why children are born to create, and what happens when you give them the tool to express it. - AI in the Passenger Seat: How Kindlewood's writing coach asks open-ended questions instead of writing the story, keeping kids as the authors. - The Persona Split: Why the reading app and the creator experience had to be completely separate products, a lesson from watching her three-year-old. - From Microsoft PM to Founder: What enterprise product thinking transfers to a kids' startup, and what you have to unlearn. - Demanding Customers Are the Best Customers: The PM lesson that the hardest customers to win become the most loyal. - Steady Over Viral: Why she's deliberately growing slow, with 9 school partnerships and workshops where kids don't want to leave. - "Is It Me the Reason You Got a Job?": The five-year-old's question that sums up the whole company. 🎧 Full episode on all podcast platforms 💬 Should kids use AI to create, or does it risk replacing their creativity? Let us know in the comments! 🔔 Please like and subscribe! Every subscriber helps our channel grow. #DN #EdTech #KidsLearning #FeifeiQiu #Kindlewood #Storytelling #AIinEducation #Startups #Parenting #Creativity Timestamps: 0:00 Intro 0:51 Why build for the youngest learners 2:30 From China to Seattle to Microsoft 3:55 Big tech vs founding: what actually changed 7:12 Product thinking she carried from Microsoft 11:29 What you have to unlearn after big tech 14:12 Why the name Kindlewood 16:16 The dragon story that started everything 21:04 Traction: 200+ young creators, 9 school partnerships 24:37 Ten years of overnight success 26:04 The viral loop of imagination 28:07 AI in education: kids stay in the driver's seat 31:36 How vibe coding changed her builder workflow 34:25 "Is it me the reason you got a job?"

  7. Aug 4

    DN #28: Runtime Economics, The AI CFO Gap & Why Expensive Models Are Cheaper (w/ Joakim Hauge)

    "It's cheaper to upgrade to an expensive model." I talk to Joakim William Hauge, co-founder of Monetize, about "runtime economics" (the new category he's carving out for AI governance) and why CFOs are locked out of the AI dashboards their engineers use every day. Joakim spent years in quantitative finance and risk management before turning that framework onto the non-deterministic runtime behavior of autonomous AI systems. He's building Unitflow, an enterprise governance layer launching soon, and already shipped Circuit Breaker on GitHub as a validation vehicle. The wedge: use runtime telemetry to translate token spend into income-statement language the CFO can actually act on. In this episode: - Runtime Economics: The new category Joakim is defining, using runtime telemetry to understand AI value creation and risks, not just cost optimization. - The CFO Gap: Why the executives signing off on AI budgets are locked out of the dashboards their engineers actually use. - Why Expensive Models Are Cheaper: The counter-intuitive result of testing, that upgrading to a more expensive model can reduce total spend by killing recursive retry loops. - Cost, Value, Risk: The three-pillar framework for governing AI spend, and why CFOs need income-statement language, not token dashboards. - Circuit Breaker: The open-source runtime governance tool Joakim shipped to validate the thesis before Unitflow's full launch. - The Uber Cost Shock: How cost-cutting headlines from big companies changed the tone of AI adoption in a matter of months. - The Usage-Based Pricing Trap: Why real-time pricing for non-deterministic AI runtime is one of the hardest unsolved problems in SaaS. - Understand the Customer's Story: Joakim's biggest founder lesson, that the shortcut isn't shipping fast, it's understanding what the buyer actually needs. 🎧 Full episode on all podcast platforms 💬 Should CFOs get their own AI dashboards, or is token spend just an engineering-team problem? Let us know in the comments! 🔔 Please like and subscribe! Every subscriber helps our channel grow. #DN #RuntimeEconomics #AIGovernance #JoakimHauge #Unitflow #AICostManagement #Startups #AIAgents #EnterpriseAI #FinTech Timestamps: 0:00 Intro 0:40 From quant finance and risk management to AI governance 2:19 Why he's building in public 5:32 The engineer vs CFO stakeholder gap in AI adoption 7:56 Runtime economics defined in one sentence 8:20 The Uber cost shock: why April changed everything 14:06 Volatility in AI cost and how to price it 19:34 What Unitflow actually does: the three pillars 23:54 Why CFOs are locked out of the AI dashboards 25:11 Frontier vs cheaper models: when to use each 27:09 Why expensive models can actually be cheaper 27:44 The GitHub Circuit Breaker tool 30:43 Founder lesson: understand the customer's story 34:04 The unsolved problem of real-time usage-based pricing

  8. Jul 28

    DN #27: Agents as Software, Small Business AI & Building in Public (w/ Julian Fitzpatrick)

    "Software will be removed. Agents are gonna be what you sell." I talk to Julian Fitzpatrick, founder of Build with Pixel, about why packages of AI agents will replace B2B software for small businesses, and the bottleneck-and-propagate playbook he uses to build tools one company at a time. Julian spent 17 years in visual effects before pivoting into AI when ChatGPT 3 dropped, and now runs a two-pronged model: consulting local small businesses to automate their workflows, then turning those tools into a public platform. We get into the volleyball academy case study, why he thinks the gap between frontier and open-weight models is about to disappear, the three-layer human-in-the-loop workflow for shipping agent code safely, and why he's building everything in public. In this episode: - Agents as the New Software: Why Julian thinks packages of agents will replace B2B software, tailored per vertical instead of per feature. - The Bottleneck-and-Propagate Playbook: How he wins one small business (a volleyball academy, an architecture firm), automates a workflow, then propagates the tool across the vertical. - Local Models Closing the Gap: Why open-weight models are approaching frontier quality, and what that means for on-prem SMB deployments. - The Three-Layer Approval Workflow: Screenshots, staging, production. The human-in-the-loop pattern that lets agents ship code without breaking things. - Building in Public as a Solo Founder: Why Julian shares everything he builds, and how that becomes trust-building distribution. - The AI Operator Role: The new job category emerging as agents get real — humans as orchestrators, not executors. - From VFX to AI Agents: 17 years in visual effects, then a pivot into agent infrastructure when ChatGPT 3 dropped. - Signal vs Noise: The Kevin O'Leary / Steve Jobs framing Julian uses to stay focused as a solo builder. 🎧 Full episode on all podcast platforms 💬 Will "packages of agents" actually replace B2B SaaS, or just add a new layer on top of it? Let us know in the comments! 🔔 Please like and subscribe! Every subscriber helps our channel grow. #DN #AIAgents #SmallBusinessAI #JulianFitzpatrick #BuildWithPixel #BuildingInPublic #Startups #SMB #AI #AgenticAI Timestamps: 0:00 Intro 0:49 What Build with Pixel is and who it serves 2:00 From 17 years in VFX to building AI agents 3:28 Open-weight models closing the gap on frontier 5:12 The local-vs-cloud choice for small business AI 8:27 The bottleneck-and-propagate methodology 9:35 Case study: the volleyball academy CTA workflow 11:43 Which industries are easiest to enter with AI now 14:17 Why he's building in public 17:06 Attention is all you need, audience as the moat 22:49 Why he doesn't obsess about ICP or revenue ranges 25:49 Steve Jobs, signal vs noise, and the solo founder trap 26:20 The 12-month arc from prompts to harnesses to agents-as-software 30:19 Human-in-the-loop and the three-layer approval workflow 33:44 Where to find Julian and Build with Pixel

See All (34)

About

Welcome to Dr. Niklas: Venture Grade - the intersection of deep tech and high-stakes business. I’m Dr. Niklas, an engineer, founder, and researcher. I look at business and technology through the lens of a builder: how do you architect a company that is robust, scalable, and "Venture Grade"? On this channel, I move beyond surface-level startup advice. I sit down with world-class founders, investors, and technical leaders to debug their strategies and examine the source code of their success. What is "Venture Grade"? It’s a standard. Just like military-grade hardware or enterprise-grade software, a Venture Grade business is built to survive, scale, and secure investment. New episodes ship every Tuesday. Subscribe for weekly deep dives with people building the future 🚀☘️

Information