Built This Week

A Bot Emailed a Sold Out Hotel for Me. 15 Minutes Later I Got 33% Off.This week there was no guest. Sam and Jordan spent the episode on bots: what they are, why they are different from chatbots and coding assistants, and how we are using them every day.The big change is that bots like Grok Bot, Muse, and OpenAI Dots run on their own browser and their own computer. They stay logged in, run workflows on a schedule, and keep working when your laptop is closed. Jordan told the story of a hotel that showed as sold out. His bot found the hotel's email, wrote to them, and 15 minutes later the hotel called back with 33% off, about $200 saved for one night.We also covered how we use agents to update our websites, from edits in Webflow to a custom JavaScript tracker with heat maps and AI generated ads, plus the latest Nvidia headline out of China.We cover:What bots are and how they differ from chatbots, sidebar assistants, and coding agentsWhy a bot with its own browser and computer is a step changeConnecting email and LinkedIn so a bot can send, reply, and make connections for youThe hotel story: 33% off and $200 savedWhy bots could disrupt aggregators like Booking.com and ExpediaUse cases: travel, big purchases like a used boat or car, social media, market research, and websitesWhy a bot's own virtual machine matters for long running workflowsHow we use agents to update our websites and build adsThe Nvidia news out of ChinaIf you are curious about bots and agents, or wondering how they will reach the enterprise, this episode is for you.New episodes every Friday at BuiltThisWeek.comTIMESTAMPS[00:00] Intro[01:16] What bots are and how they differ from chatbots and coding assistants[02:16] The step change: bots with their own browser and computer[02:52] Connecting email and LinkedIn to your bot[03:51] New workflows for sales and the enterprise[05:03] Sam on three weeks of using Grok Bot[05:45] The hotel story: 33% off and $200 saved[06:38] Why bots could disrupt aggregators like Booking.com and Expedia[07:16] More use cases: big purchases, social media, research, websites[08:08] Why a bot's own virtual machine changes everything[09:27] Using agents to update our websites[10:41] News: Nvidia and China[11:19] Wrap up#AI #AIAgents #Bots #GrokBot #MetaMuse #OpenAI #Nvidia #BuildWithAI #BuiltThisWeek

This week we sat down with Ivan Lee, co-founder and CEO of Datasaur, to dig into why the most regulated companies on earth are running AI on their own servers. Ivan worked on AI products at Yahoo and then on Siri at Apple before founding Datasaur eight years ago. The FBI, Google, Netflix, Zoom, Deloitte, and AmLaw 100 law firms are all clients. Greg Brockman, co-founder of OpenAI, is an investor.We covered the real token economics of enterprise AI, why the open weights model performance gap has almost closed, how Datasaur's proprietary model factory works, and Ivan's personal MCP build connecting Claude to his Whoop bracelet for a daily training plan.We also covered Meta Muse crossing 10 million users in 6 weeks, Anthropic launching Claude Duet, and why everyone is underestimating Google's personal agent.We cover:What Datasaur does — private on-premise AI for the most regulated enterprises on earthWhy companies are switching from public AI after Anthropic launched its own legal and design productsHow the performance gap between proprietary and open weights models has almost closedMeta's 100 billion parameter model running on two Mac Studios — and what that means for private AI costsOne AI agent task uses 3 million tokens vs 2,000 for a ChatGPT queryAI agents crossing the chasm — what happens to token demand when they doDatasaur's model factory — white glove consulting on the outside, productized AI factory on the insideIvan building an MCP server for his Whoop bracelet and giving Claude control of his training planWhy enterprise agents are a better fit than consumer agents right nowMeta Muse crossing 10 million users in 6 weeks with shopping as the top activityAnthropic launching Claude Duet — and why Ivan, who worked on Siri, is watching closelyThe three-way race between OpenAI Dots, Meta Muse, and Anthropic DuetWhy Google's personal agent is being underestimated by everyoneIf you are building with AI, managing enterprise data, or want to understand the real economics of the agent wave, this episode is for you.Find Ivan and Datasaur:Website: datasaur.aiLinkedIn: linkedin.com/in/iyliTwitter: x.com/datasauraiNew episodes every Friday at BuiltThisWeek.comTIMESTAMPS[00:00] Intro[00:42] Meet Ivan — co-founder and CEO of Datasaur[01:00] Yahoo, Apple Siri, and the path to founding Datasaur[01:24] What Datasaur builds — private AI inside your own infrastructure[02:37] Healthcare, legal, finance, and public sector — who uses it[03:01] Why regulated industries are moving faster on AI than any technology in history[03:54] Domain expertise as the moat — Anthropic enters clients' markets[04:53] Open weights models catching up — the performance gap is closing[05:53] Live benchmark — every open model now clears Claude 4.5[07:24] Hardware costs plummeting — 100B parameter model on two Mac Studios[08:57] The benchmark slide — where agents started actually working[10:43] One AI task uses 3 million tokens — the macro token shift[11:39] Crossing the chasm — chatbots vs agents on the adoption curve[13:41] Agents collapsing software spend — the CPU to token shift[14:23] How Datasaur works with clients — the model factory[16:36] News — Meta Muse crosses 10 million users[17:26] Ivan worked on Siri at Apple — personal agents have always been his passion[18:09] The consumer agent gap — why it took so long[19:14] Jordan on Grok Bot and underestimating Google[20:15] Enterprise vs consumer agents — where the market fit is now[21:08] Ivan's favorite AI build — MCP server for his Whoop bracelet[22:14] Where to find Ivan and Datasaur[22:42] Wrap up#AI #PrivateAI #Datasaur #Enterprise #AppleSiri #OpenAI #BuildWithAI #BuiltThisWeek #Anthropic #MetaMuse #ClaudeDuet #AIAgents

This week we sat down with Jim Olsen, co-founder and CTO of ModelOp, to explore what it takes to get AI into production inside Fortune 500 companies. Building a model is only part of the job. Knowing where it runs, what it costs, who owns it, and whether it meets company policies is where things get complicated.Jim walks us through a live demo of ModelOp, showing how it discovers unregistered enterprise AI, brings it into a central inventory, and uses agents to fill out documentation and flag compliance gaps. His team is turning work that can take months into a process that takes minutes, with humans still in the loop.We also covered why AI spending needs to be measured by business use case, whether an AI-powered pet feeder solves a real problem, and how Jim runs engineering from an off-grid home at 10,200 feet in the Colorado mountains.We cover:• What ModelOp does and why all its clients are Fortune 500 companies• Why enterprise AI projects get stuck between prototypes and production• Tracking token usage and costs by business use case• Shadow AI beyond employees using chatbots• A live demo of discovering and registering enterprise AI• Using AI agents to complete forms and identify compliance gaps• Why model inventories matter when providers change or models retire• The real-world problems that led Jim to co-found ModelOp• Petlibro’s AI-powered feeder and whether everything needs AI• Running an engineering team with solar power and Starlink from an off-grid homeIf you are building with AI, managing enterprise AI deployments, or trying to understand how governance can accelerate delivery, this episode is for you.Find Jim and ModelOp:Website: https://www.modelop.comNew episodes every Friday at BuiltThisWeek.comTIMESTAMPS[00:00] Intro[00:41] Meet Jim Olsen, co-founder and CTO of ModelOp[01:49] What ModelOp does: enterprise AI governance explained[03:15] Tracking AI costs by business use case[04:50] Demo: token usage and cost dashboards[05:41] Automating governance and compliance workflows[06:08] Discovering shadow AI across enterprise platforms[08:49] Bringing a healthcare chatbot into the inventory[09:56] AI agents automate documentation and forms[11:20] Checking policies and flagging compliance gaps[12:05] From months of work to minutes[12:27] The real-world problems behind ModelOp[14:04] Tracking model dependencies and managing changes[15:23] Where to find ModelOp[15:43] News: Petlibro’s AI-powered pet feeder[16:11] Does everything really need AI?[17:20] Automatic feeders, robots, and everyday chores[18:23] Jim’s off-grid setup in the Colorado mountains[19:09] Solar panels, batteries, and working remotely[20:12] Wrap-up#AI #ModelOp #AIGovernance #EnterpriseAI #ShadowAI #AIAgents #Fortune500 #BuildWithAI #BuiltThisWeek

This week we sat down with Rob Williams, co-founder and CTO of Read AI, to dig into what 5 million monthly active users actually looks like at the product and infrastructure level. Most people know Read AI as a meeting note taker. Rob winces every time he hears that. What it really does is capture every conversation across your organization and turn it into intelligence — daily briefings, smart routing, and organizational context that moves itself to where it needs to be.We also covered AI companion apps and the unit economics behind them, the AI adoption divide inside enterprise companies, and Opus 5.5 dropping this week.We cover:What Read AI actually is beyond the meeting note taker labelThe daily briefing feature — hot and hydrated context before every meetingHow smart routing sends meeting data automatically to the right teamWhy some companies set data retention to 7 days and why that makes senseNavigating the AI adoption divide inside enterprise organizationsRob's vision for information moving itself in three yearsWhy meetings are the biggest investment most companies make and nobody measuresHot Girl Hotline — an AI relationship advice app and why the unit economics workAI as a sounding board vs a solver and why both have valueOpus 5.5 dropping and what cheaper intelligence means for buildersToken economics of running 5 million monthly active usersIf you are building with AI, managing a team, or want to understand where enterprise AI is actually heading, this episode is for you.Find Rob and Read AI:Website: readai.comEmail: [email protected]: [LINKEDIN LINK — to be added]New episodes every Friday at BuiltThisWeek.comTIMESTAMPS[00:00] Intro[01:06] Meet Rob — co-founder and CTO of Read AI[01:42] What Read AI actually is beyond meeting notes[02:12] Meetings as the biggest investment companies make[02:33] The daily briefing feature — live demo[03:47] Smart routing — data flows where it needs to go[04:57] Unusual use cases and the AI adoption divide inside companies[06:19] 7-day data retention and what it says about fast-moving companies[06:40] 5 million monthly active users and how they got there[07:00] Token economics at scale[10:00] The future — information that moves itself[15:34] What supersedes old context — the data freshness problem[15:55] News — Hot Girl Hotline, AI relationship advice[17:09] Is AI as a companion dangerous or valuable[18:14] AI memory and the sounding board use case[19:37] Unit economics of AI companion products[20:13] Sam uses AI for big personal decisions[21:32] Where to find Rob[21:50] Wrap up#AI #ReadAI #Meetings #Enterprise #FutureOfWork #BuildWithAI #BuiltThisWeek #Productivity #Anthropic #AICompanion

He Ran a Rocket Launch Company. Now He Is Building Autocomplete for Hardware Design.This week we sat down with Sohrab Haghighat, CEO and co-founder of Hestus, to demo the AI layer that sits on top of your existing CAD software and autocompletes your next design move. Rob spent years as an aerospace engineer, ran a rocket launch company, and built Hestus to eliminate the manual clicks that slow hardware engineers down and introduce errors into designs.We also covered Anthropic 5.5, GPT-6, and a new Grok model all dropping in the same week — and why intelligence getting cheaper is good news for everyone building right now.We cover:What Hestus is and how AI autocomplete works inside CAD softwareLive demo — AI suggesting the next design step before you clickWhy Hestus plugs into your existing CAD tool instead of replacing itHow Hestus learns your preferences within a design sessionWho the target customer is — mechanical, aerospace, and robotics engineersThe gap between a flashy 3D model and one that is actually ready for manufacturingRob's long term vision — doing for hardware what Claude Code did for softwareAnthropic 5.5, GPT-6, and Grok all dropping the same week — intelligence is getting cheaperWhy LLM improvements do not directly translate to improvements in hardware AIIf you are building hardware, working in engineering, or want to understand where AI is going beyond software, this episode is for you.Find Rob and Hestus:Website: hestus.coLinkedIn: linkedin.com/in/sohrabhaghighatTwitter: x.com/s_haghighNew episodes every Friday at BuiltThisWeek.comTIMESTAMPS[00:00] Intro[01:12] Meet Rob — CEO and co-founder of Hestus[01:33] Aerospace engineer, rocket launch company, and the pain that led to Hestus[02:29] Live demo — AI autocomplete inside CAD software[03:50] How Hestus sits on top of your existing CAD tool[04:16] Meeting engineers where they are — no new workflow[05:02] How Hestus learns your preferences within a session[06:32] Available on Autodesk Fusion now, SolidWorks coming soon[07:20] Who is using Hestus — mechanical, aerospace, robotics engineers[08:30] Bottom-up approach — sketch to 3D to manufacturing readiness[10:25] Long term vision — Claude Code for hardware design[11:15] The gap between a flashy model and one ready for manufacturing[12:00] News — Anthropic 5.5, GPT-6, and Grok released same week[13:18] Jordan tests the new Opus — cheaper and competitive with Fable[14:41] Intelligence is getting cheaper — what that means for builders[15:06] Why LLM improvements do not directly improve Hestus[16:08] Where to find Rob and Hestus[16:31] Wrap up#AI #CAD #HardwareDesign #Engineering #Hestus #BuildWithAI #BuiltThisWeek #Anthropic #OpenAI #Aerospace

This week we sat down with Jennifer Dulski, CEO and founder of Rising Team, to demo the AI leadership platform that acts as a personal trainer for your effectiveness at work. Jen spent 10 years at Yahoo, then Google, then Facebook, ran Change.org for five years, and sold her first AI company to Google in 2009. That company now powers the review insights inside Google Maps. She also teaches at Stanford GSB.Rising Team's AI coach RT does not give generic leadership advice. It knows your team. It knows that Tom likes direct feedback and Debbie needs encouragement first. And it helps you practice the conversation before you have it.We also got into why Apple Intelligence took too long, why AI memory still feels creepy to some people, and why the best AI features are the ones you never notice.We cover:What Rising Team is and how the AI coach RT worksWhy 70 percent of AI transformations fail because of people not technologyHow RT learns your team individually and personalizes every conversationWhy every executive gets a coach and every frontline manager gets nothingWhy the engagement survey is dying and what is replacing it in real timeThe privacy guardrails built into RT and EU AI Act complianceJen's first AI company in 2009 — DealMap, sold to Google, now inside Google MapsWhy Apple Intelligence took so long and whether it matters anymoreThe AI memory debate — helpful or creepyWhy the best AI features are the ones you never have to turn onIf you are building with AI, managing a team, or want to understand where workplace AI is heading, this episode is for you.Find Jen and Rising Team:Website: risingteam.comLinkedIn: linkedin.com/in/jdulskiNew episodes every Friday at BuiltThisWeek.comTIMESTAMPS[00:00] Intro[01:11] Meet Jen — CEO and founder of Rising Team[01:44] From high school teacher to Yahoo, Google, Facebook, and Stanford GSB[02:27] What Rising Team is at the highest level[02:47] Why coaching never scaled past the executive level[03:40] How Rising Team started in 2020[04:03] DealMap — her first AI company, sold to Google, now inside Google Maps[05:18] How Rising Team evolved from team building software to AI coach[05:55] How RT knows your team individually[06:31] Live demo — RT in action[07:00] Why 70 percent of AI transformations fail because of people[07:24] The science of high performing teams has not changed[08:12] What a session with RT actually looks like[10:00] Real-time anonymized insights for executives[14:21] Why the engagement survey is dying[14:48] Real insights from enterprise customers[15:36] Privacy guardrails and EU AI Act compliance[16:43] News — Apple Intelligence, Siri, and why it took too long[18:28] The AI memory debate[20:06] The best AI features are the ones you never notice[21:00] Where to find Jen and Rising Team[21:42] Wrap up#AI #Leadership #RisingTeam #FutureOfWork #BuildWithAI #BuiltThisWeek #TeamManagement #AICoach #Apple #HR

This week we sat down with David Petrou, CEO of Continua, to demo Wheely — an agent development operating system built for developers running multiple AI coding agents who are hitting the wall of mental overhead. David spent 18 years at Google, was on the founding teams of Google Glass and Google Goggles, and left when ChatGPT launched. We also covered the foldable iPhone, Android vs Apple for AI agents, Fable 5.1, Anthropic vs OpenAI on developer experience, and whether new programming languages designed for agents are coming.We cover:What Wheely is and why David calls it an agent development operating system18 years at Google, Google Glass, Google Goggles, and why he left to start a companyThe pivot from social AI to building the OS layer for multi-agent codingWhy token costs dropping does not solve the human attention problemHow Wheely decomposes problems, allocates agents, and knows when to interrupt youWhy Git and GitHub were designed for humans and need to be rethought for agentsHow agent coding can modernize legacy code and rebuild bloated React frontendsWho Wheely is actually built for and why the middle space is the opportunityWhy new programming languages designed specifically for agents could unlock major efficiency gainsApple vs Android for AI — best hardware vs best OS for agent use casesHow Android can disintermediate apps by running them simultaneously off screenFable 5.1, Gemini 3.8 Flash, and the honest developer take on Anthropic vs OpenAIThe coevolution of models, harnesses, and programming languagesWhether the foldable iPhone matters if agent coding eliminates the need for extra screen real estateIf you are building with AI agents, working in dev tools, or want to understand where the agent coding stack is heading, this episode is for you.Find David and Wheely:Website: wheelie.devTwitter: x.com/dpetrouNew episodes every Friday at BuiltThisWeek.comTIMESTAMPS[00:00] Intro[01:07] Meet David — CEO of Continua and creator of Wheely[01:56] Google Glass, Google Goggles, 18 years at Google[02:40] Why he left Big Tech when ChatGPT launched[05:18] The previous product — social AI for group chats[06:36] When coding agents changed everything[08:03] How building for agents internally became the real opportunity[08:56] Who Wheely is built for[09:46] What happens when you run 4 agents at once[10:17] Human attention is the scarcest resource in agent coding[11:16] How Wheely works — decompose, allocate, optimize, interrupt only when needed[13:05] The space between Claude Code and Lovable[14:05] Modernizing legacy code and greenfield agent projects[15:18] Why React is bloated and agents can rebuild frontends better[16:11] News — Apple foldable iPhone coming Q3[17:29] Apple unified memory vs Android open OS for AI agents[18:27] Android disintermediating apps with off-screen agent execution[20:24] Will the foldable iPhone matter if agents take over development[20:48] Fable 5.1 launch — honest developer take[22:10] Anthropic vs OpenAI on developer experience and harness access[23:17] New programming languages designed for agents[25:19] The coevolution of models, harnesses, and programming languages[25:40] Where to find David and Wheely[25:55] Wrap up#AI #AIAgents #Wheely #DevTools #GoogleGlass #BuildWithAI #BuiltThisWeek #Anthropic #Fable #Apple #iPhone

This week we sat down with Hardik Kabaria, co-founder and CEO of Vinci, to demo the AI platform making physics simulation accessible to any engineer building hardware. A chip tape-out used to cost millions of dollars and take months. Thermal analysis that should take seconds was waiting days for a specialist with a PhD. Vinci is compressing all of that into minutes — and making it accessible to anyone on the design team, not just the simulation expert.We also got into why AI image and video generation still has not produced a Netflix movie, what has to change before it does, and how AI getting into hardware design raises the stakes around IP and watermarking in ways nobody is talking about yet.We cover:What Vinci does — making physics simulation accessible to anyone building hardwareWhy a chip tape-out costs millions and takes months and how AI is compressing that cycleHow Vinci lets engineers get thermal and physics answers in seconds instead of daysThe existing workflow — designers, performance evaluators, and manufacturers rarely even work in the same teamHow simulation software went from requiring PhDs to something any engineer can accessThe digital twin concept — running physics experiments in software before touching physical prototypesWhy the semiconductor and data center cooling market is the starting pointNews — AI image and video generation still has not crossed the quality schismWhy there are no AI-generated movies on Netflix yet even though the models existHow AI getting into hardware design changes the stakes around IP and watermarkingThe parallel between language models democratizing reasoning and physics models democratizing simulationIf you are building hardware, working in semiconductors, or want to understand where AI is going beyond software, this episode is for you.Find Hardik:LinkedIn: linkedin.com/in/hardikkabariaTwitter: x.com/hardikk13Find Vinci:Website: getvinci.aiLinkedIn: linkedin.com/company/vinciphysicsCompany Twitter: x.com/VinciPhysicsNew episodes every Friday at BuiltThisWeek.comTIMESTAMPS[00:00] Intro[01:14] Meet Hardik — co-founder and CEO of Vinci[02:06] What Vinci does — making physics accessible for hardware builders[03:24] How simulation has worked in hardware for decades[04:33] The digital and physical world of experiments[05:47] Getting physics answers in seconds instead of days[06:28] How Vinci fits into the existing hardware design workflow[08:10] The tape-out problem — millions of dollars and months per iteration[10:00] Live demo — thermal simulation for semiconductor design[13:00] Who is using Vinci and what problems they are solving[15:30] The language model analogy — democratizing physics reasoning[17:00] News — AI image generation and the quality schism[19:30] Why there are no AI movies on Netflix yet[21:00] AI getting into hardware and the IP watermarking problem[23:50] The regulations and guardrails that will follow[26:05] Where to find Hardik and Vinci[26:43] Wrap up#AI #Physics #Semiconductors #HardwareDesign #Vinci #ChipDesign #BuildWithAI #BuiltThisWeek #AIHardware #Simulation

This week we sit down with Zach Rattner, CTO and co-founder of Yembo, to demo the computer vision platform that turns a ten minute phone walkthrough into a full home inventory, 3D model, and floor plan. Deployed in 35 countries, running thousands of inspections every day for moving companies and property insurers — and the end user never even knows AI is involved. We also dig into Claude Fable 5 being pulled without coding access, the 25,000 Chinese accounts exploiting the model, and whether AI has finally entered its regulatory adolescence.We cover:What Yembo does — computer vision that deeply understands the interior of homesHow a ten minute phone video replaces a 60 to 90 minute in-home inspectionThe visual inventory — color coded, numbered, time stamped photos of every itemWhy the real AI opportunity is doing what was previously impossible, not just fasterHow Yembo embedded AI in a workflow where the end user never knows it existsThe insurance use case — 3D model and floor plan from a single video walkthroughWhy 30 percent of their patents are on guiding the camera capture without any trainingClaude Fable 5 pulled and reinstated without coding access — what it means for developersAI reaching regulatory adolescence and whether the framework arrives in time25,000 Chinese accounts exploiting the model and the KYC failure behind itIf you are building with AI, working in insurance or logistics, or want an honest take on what happens when regulators start touching frontier models, this episode is for you.Find Zach and Yembo:Website: yembo.aiLinkedIn: search Zach RattnerNew episodes every Friday at BuiltThisWeek.comTIMESTAMPS[00:00] Intro[01:15] Meet Zach — CTO and co-founder of Yembo[02:05] What Yembo does — computer vision for home interiors[03:30] How a ten minute phone video replaces 90 minutes of in-home inspection[05:10] Live demo — the visual inventory report[07:00] The real AI opportunity — doing what was previously impossible[09:30] Embedding AI so the end user never knows it exists[11:00] Insurance use case — 3D model and floor plan from video[13:00] 30 percent of patents on guiding camera capture without a tutorial[14:30] News — Claude Fable 5 pulled and reinstated without coding[16:20] AI adolescence — the regulation debate[18:30] 25,000 Chinese accounts and the KYC problem[20:30] Where to find Zach and Yembo[21:00] Wrap up#AI #ComputerVision #Yembo #Insurance #MovingIndustry #ClaudeFable #BuildWithAI #BuiltThisWeek #PropTech #AIRegulation

This week we sit down with Ryan Aytay, President and COO of Code Metal, to dig into one of the most underestimated problems in tech — the legacy code running mission critical infrastructure across aerospace, defense, automotive, and medical devices. Code Metal translates that code to any hardware with mathematical proof it will behave identically. Provably correct. We also dig into Meta glasses, Jony Ive at OpenAI, and where AI hardware interfaces are really heading.We cover:What Code Metal does and why provably correct code mattersThe legacy code problem running mission critical infrastructure right nowWhy testing is not enough and what formal methods actually proveHow AI finally made formal verification scalable after 50 yearsThe Tesla over-the-air update analogy for code-to-hardware deploymentWhich industries Code Metal is built for and whyWhy vibe coding works for apps but terrifies Ryan for aerospace and defenseWhat brought Ryan from running Tableau at Salesforce to a code verification startupMeta Ray-Ban glasses — price drop, Kylie Jenner, and the veteran use caseJony Ive at OpenAI and what the mystery hardware device might beWhy every AI interface is converging on the same form factorIf you build software, work in hardware, or want to understand the infrastructure layer AI is about to touch, this episode is for you.Find Ryan and Code Metal:Website: codemetal.aiLinkedIn: linkedin.com/in/ryanaytayCompany LinkedIn: linkedin.com/company/code-metalTwitter: x.com/Code_Metal_AITIMESTAMPS[00:00] Intro[01:04] Meet Ryan — President and COO of Code Metal[02:00] What Code Metal does in one sentence[03:30] The legacy code problem in mission critical systems[05:32] What provably correct means and why testing is not enough[07:10] Formal methods — mathematical proof not just testing[08:45] How AI made formal verification scalable[10:20] Tesla over-the-air updates analogy[11:40] ICP — aerospace, defense, automotive, medical[13:05] Why vibe coding terrifies Ryan for mission critical systems[15:10] From Salesforce and Tableau to Code Metal — the origin story[17:30] News — Meta Ray-Ban glasses price drop and Kylie Jenner[19:45] Meta glasses given to veterans — the real use case[20:50] Jony Ive at OpenAI and the mystery hardware device[22:30] Why all AI interfaces are converging on glasses[23:05] Where to find Ryan and Code Metal[23:45] Wrap upNew episodes every Friday at BuiltThisWeek.com#AI #CodeMetal #FormalMethods #LegacyCode #Engineering #Defense #Aerospace #BuildWithAI #BuiltThisWeek #AIHardware