Built This Week

Training gets the headlines.Inference is where the money is.In Episode 37 of Built This Week, we sit down with Mitesh, CEO of Positron AI, to break down one of the biggest bottlenecks in AI today: inference infrastructure.While the world focuses on trillion-parameter models and frontier labs, the real constraint isn’t intelligence — it’s memory, bandwidth, energy, and cost.We cover:• Why inference is where 90% of AI spend happens • The memory wall problem in large models • Why GPUs weren’t designed for text generation • How Positron is building terabyte-plus memory chips • The economics of 10 trillion parameter models • Why memory bandwidth utilization matters • Why CPUs are suddenly back in demand • The difference between speed-optimized and cost-optimized AI systems • The slider bar future of AI infrastructureWe also dive into:• OpenAI’s $122B valuation • Anthropic vs OpenAI secondary market dynamics • Why Nvidia isn’t going anywhere • Why commodity memory might beat premium stacks in certain use cases • The rise of agentic workflows and what that means for computeIf you care about the future of AI, silicon, infrastructure, or trillion-dollar companies — this episode is for you.New episodes every Friday.⏱ TIMESTAMPS(0:00) Why inference is the real AI bottleneck (2:00) What Positron AI is building (4:30) The memory problem in trillion-parameter models (6:30) Why GPUs struggle with inference economics (9:00) Energy, bandwidth, and supply chain constraints (12:00) Memory capacity vs memory speed tradeoffs (16:00) The “slider bar” model of AI infrastructure (18:30) OpenAI’s $122B valuation discussion (21:00) Anthropic vs OpenAI secondary markets (23:30) CPUs making a comeback (26:00) Agentic workflows and compute demand explosion (28:00) Closing thoughts on AI infrastructure

Our recruiters are not video editors.But now they can cut highlight reels in minutes.In Episode 36 of Built This Week, we break down a tool we built internally at Ryz Labs that lets our recruiting team generate polished candidate highlight videos without touching Premiere, Final Cut, or CapCut.The problem:When presenting candidates to clients, resumes are standard. But seeing a candidate speak for 60 seconds changes everything.The issue was speed. Editing sizzle reels required our video team, added delays, and was not scalable.So we built a highlight reel generator powered by:• EntreVista AI interview transcripts • Claude and Codex for clip selection • Remotion for video rendering via code • AWS S3 for instant share linksThe system automatically: • Analyzes transcripts • Identifies high signal clips • Groups them by communication, role fit, and personality • Allows light manual adjustments • Renders a branded video in 5 to 10 minutesNo editing experience required.Then we dive into Remotion and why “video as code” is one of the most underrated AI enabled workflows right now.Finally, we discuss the growing cost of AI usage inside organizations: • Token spend management • Surprise AI bills • Model access guardrails • Productivity vs cost tradeoffsAI is democratizing building.But it is also introducing a new management layer.New episodes every Friday.⏱ TIMESTAMPS(0:00) The problem: recruiters are not video editors (0:25) Welcome to Episode 36 (1:20) Why highlight reels improve candidate selection (2:30) The scalability issue with manual video editing (3:30) Demo: AI Highlight Reel Builder (4:15) How transcripts power automatic clip selection (5:00) Communication, role fit, personality grouping (6:10) Manual adjustments for recruiters (7:00) Rendering time and infrastructure challenges (8:00) Final sizzle reel output demo (9:00) How it was built with Codex (10:00) What is Remotion (11:30) Video editing as code explained (12:30) Other Remotion use cases: product trailers, documentation videos (13:45) Democratizing creative production (14:30) AI token costs inside organizations (15:15) Surprise AI bills and infrastructure lessons (16:30) Managing model access across teams (17:30) Productivity vs spend tradeoffs (18:15) Closing thoughts🔗 LINKSBuilt This Week New episodes every Friday https://builtthisweek.comJordan Metzner https://x.com/mrjmetzSam Nadler https://x.com/Gravino05

Marketing teams are about to change forever.Instead of hiring designers, copywriters, analysts, SEO specialists, and performance marketers… companies are starting to run AI marketing agents that handle everything.From planning campaigns to creating content, analyzing performance data, generating ads, and optimizing strategy automatically.In this episode of Built This Week, Sam Nadler and Jordan Metzner sit down with Iliya Valchanov, CEO of Juma, to explore how AI agents are transforming modern marketing workflows.Juma is building an AI marketing super-agent that can autonomously plan, execute, and optimize marketing campaigns across channels like social media, ads, analytics, and SEO.During the episode, Ilia demos how a single prompt can generate a complete social media strategy, content calendar, and visual assets in minutes.Even more surprising — Ilia explains how their own company replaced a 7-person go-to-market team with just one person using AI agents.We also dive into the future of AI agents, how developers are working with coding agents like Claude Code and Codex, and why the next wave of AI tools may turn websites into constantly evolving, self-optimizing systems.In this episode we discuss:• How AI agents automate entire marketing workflows • Turning a single prompt into a full social media calendar • Why marketing teams are early adopters of AI • How AI agents connect to tools like Google Analytics, HubSpot, and Meta Ads • Why saving time isn’t the real benefit of AI • How AI increases marketing quality and experimentation • How agencies measure ROI and billable hour savings from AI • Why companies need a dedicated “AI transformation leader” • Claude Code vs Codex vs Cursor for AI coding workflows • Andrej Karpathy’s new auto-research AI experimentsThis episode is a glimpse at how AI agents may reshape marketing, coding, and digital products over the next decade.⏱️ TIMESTAMPS(0:00) Welcome to Built This Week (0:31) Introducing Ilia from Juma (0:46) What Juma is building (1:26) Live demo: AI marketing agent (2:10) Generating a social media calendar with one prompt (3:02) Researching competitors automatically (3:52) Building a full content strategy (4:30) Creating Instagram carousels with AI (5:21) Integrations with Google Analytics, HubSpot, and Ads (6:03) Can AI learn which content performs best? (6:49) Who is using Juma today (7:40) Marketing teams vs marketing agencies (8:05) Replacing a 7-person marketing team with AI (8:58) Publishing blog posts in 3 minutes (9:40) Why AI unlocks new marketing opportunities (10:27) Measuring ROI and billable hours saved (11:06) How AI removes the need for specialized marketing roles (12:00) Why ad optimization is the biggest AI opportunity (12:27) Biggest lessons from running AI agents in marketing (13:04) Why companies need an AI transformation leader (14:02) Claude Code vs Codex vs Cursor (16:00) The future of AI coding agents (18:56) Why developers are reading less code (20:45) How programming may change in the AI era (22:06) Andrej Karpathy’s new auto-research AI tools (24:38) AI experiments and self-optimizing websites (27:10) Final thoughts on the future of AI agents (28:06) Where to find Juma🔗 LINKSJuma https://juma.aiBuilt This Week New episodes every Friday🎙️ HOSTSJordan Metzner https://linkedin.com/in/jordanmetzner https://x.com/mrjmetzSam Nadler https://linkedin.com/in/sam-nadler-1881b75 https://x.com/Gravino05

Clinical trials are one of the slowest and most expensive processes in modern medicine.It can take 10–15 years and up to $3 billion to bring a new drug to market — and many trials fail simply because they can’t enroll enough patients.In this episode of Built This Week, Sam Nadler and Jordan Metzner sit down with Dr. Chadi Nabhan, Chief Medical Officer at RyghtAI, to explore how AI-powered digital twins of clinical trial sites can dramatically improve the speed and success of clinical trials.RyghtAI has built a platform that creates digital twins of thousands of clinical trial sites worldwide, allowing pharmaceutical companies to instantly identify the best locations and investigators for any given trial.Instead of relying on manual site selection or reputation-based decisions, AI analyzes historical trial performance, patient demographics, biomarker capabilities, and infrastructure to determine which sites are most likely to enroll patients successfully.The result: faster trials, better patient representation, and potentially life-saving therapies reaching the market sooner.In this episode we discuss:• Why 80% of clinical trials fall behind schedule • Why half of clinical trial sites enroll 0–1 patients • How AI parses 200-page trial protocols in seconds • The role of digital twins in predicting trial success • How AI improves patient diversity in clinical trials • Why biomarker data is becoming essential in modern medicine • How AI agents infer site capabilities from historical trial data • Why informed patients using AI tools may actually improve healthcare outcomesIf AI can dramatically improve the speed and efficiency of clinical trials, it could reshape how quickly new treatments reach patients worldwide.⏱️ TIMESTAMPS(0:00) Welcome to Built This Week (0:37) Introducing Dr. Chadi Nabhan from Ryght AI (1:12) What RyghtAI is building (2:14) The problem with clinical trial site selection (3:07) Digital twins for clinical trial sites (4:01) Manual vs AI-driven trial strategy simulation (5:15) Why clinical trials fail (6:03) The massive cost and time of drug development (6:51) How AI identifies the best trial sites (8:00) Ranking clinical trial sites using AI scoring (9:03) Diversity challenges in clinical trials (10:02) Using census data to improve patient representation (10:35) Biomarkers and genomic trial requirements (11:48) Predicting future trial success from past data (12:14) How AI accelerates trial matching (13:04) AI agents reading clinical trial protocols (14:20) Parsing 200-page protocols in seconds (15:00) AI identifying investigators and site contacts (15:57) Helping overlooked clinical sites get discovered (17:47) AI’s expanding role in healthcare innovation (18:00) Eight Sleep raises $50M at a $1.5B valuation (21:09) Apple releases a $599 MacBook (23:00) Dr. Nabhan’s upcoming book: AI and Cancer Care (23:33) Will AI replace Google for patient research? (25:30) The future of personalized AI healthcare (26:10) Final thoughts and wrap-up🔗 LINKSRyght AI https://ryght.aiDr. Chadi Nabhan https://chadinabhan.comBuilt This Week New episodes every Friday🎙️ HOSTSJordan Metzner https://linkedin.com/in/jordanmetzner https://x.com/mrjmetzSam Nadler https://linkedin.com/in/sam-nadler-1881b75 https://x.com/Gravino05

Construction has lagged behind every major industry in technology adoption.Manual data entry. Spreadsheets. Email-based procurement. Slow invoice approvals. Paper delivery tickets.That’s finally changing.In this episode of Built This Week, Sam Nadler and Jordan Metzner sit down with Eldar (Field Materials AI) to break down how AI is automating procurement for commercial and civil contractors — from reading quotes and invoices to verifying pricing, matching delivery tickets, and integrating directly with ERPs.Field Materials builds AI agents that eliminate manual data entry across the procure-to-pay cycle for electrical, mechanical, concrete, drywall, and other commercial subcontractors working on hospitals, data centers, and billion-dollar infrastructure projects.We also explore:• Why construction productivity has barely improved in decades • How AI agents read and process supplier quotes automatically • How foundational model improvements upgrade products overnight • Why procurement automation directly impacts margin • The data center boom forcing construction to modernize • The difference between “adding AI” and building AI-first software • Whether incumbents like SAP and Salesforce are at risk • Why we may be entering a golden era for construction technologyThis isn’t theoretical AI.This is production AI operating inside large-scale commercial construction projects today.⏱️ TIMESTAMPS(0:00) Entering the golden era of construction tech (0:24) Welcome to Built This Week (0:43) Introducing Field Materials AI (1:12) What Field Materials actually does (1:41) Scenario modeling demo (BOM shock analysis) (3:51) Pricing intelligence and risk modeling (4:53) How the company started (6:13) Automating quotes, invoices, and delivery tickets (7:23) Who uses Field Materials (commercial subs) (8:49) How procurement actually works today (manual chaos) (10:07) Cutting overhead and scaling without hiring (11:29) Reducing material waste and pricing errors (12:25) Accelerating invoice approval cycles (13:04) AI agents for different document types (14:01) How foundational model upgrades improve the product (15:09) Why construction underinvested in tech (15:52) The data center boom forcing modernization (16:49) AI + robotics + prefabrication (17:31) Anthropic partnerships and enterprise AI integration (18:39) The next wave: AI with “hands” in enterprise systems (19:49) Why incumbents risk building gimmicks (21:07) Salesforce, SAP, and retention vs innovation (24:12) COBOL, modernization, and disruption cycles (26:39) Why building real AI tools is still hard (27:03) Where to find Field Materials🔗 LINKSField Materials https://fieldmaterials.aiBuilt This Week New episodes every Friday🎙️ HOSTSJordan Metzner https://linkedin.com/in/jordanmetzner https://x.com/mrjmetzSam Nadler https://linkedin.com/in/sam-nadler-1881b75 https://x.com/Gravino05