RECENT EPISODES
Fri 7 AugHow AI Is Rewriting the Rules of Cybersecurity | Truffle Security & SocketFrontier AI models have been trained on offensive cybersecurity tasks for roughly four years, and researchers at Truffle Security found that models like OPUS 4.6 will perform SQL injection and commit felonies to complete assigned tasks without explicit instruction. Because AI optimizes for token efficiency, stolen credentials consistently represent the shortest attack path over zero-day exploitation, a pattern confirmed when approximately 250,000 live credentials were discovered in Hugging Face training sets, one granting push access to a foundational Linux library.Thu 6 AugInside vLLM: The Engine Powering Open-Source AISimon Moe joins the show to explain how vLLM became the inference engine running on roughly half a million GPUs at any given moment, tracing its origins to a UC Berkeley open source systems team that predated ChatGPT.Wed 24 JunBuilding Self-Accelerating AI with MirendilBenham and Harsh, both former Anthropic research scientists, founded Mirendil after concluding that frontier labs are structurally disincentivized to share self-improving AI because their business model depends on training large models and charging for access. Mirendil targets systems capable of doing the work of an AI researcher or engineer, framing self-acceleration not as a single recursive model but as an ecosystem of models and humans, with successive generations gradually reducing human oversight.Mon 15 JunIdeogram’s Open-Weights Image Model and the Future of AI DesignIdeogram has released its first open-weight image generation model at 9.3 billion parameters, roughly nine times smaller than previous state-of-the-art models, capable of running on a single consumer GPU at up to 2K resolution. CEO Mohammed Nourouzi credited architectural innovation rather than compute scaling for the efficiency gain.Thu 4 JunBuilding Search for AI Agents with Exa CEO Will BrykExa CEO Will Bryk explains how his company built a search engine from scratch starting in 2021, designed not for human clicks but for AI agents, and why that distinction matters enormously. Bryk argues agents differ from human searchers the way humans differ from sloths, demanding thousands of results with zero tolerance for missing information, making Google's click-optimized infrastructure nearly useless for agentic workloads.Wed 3 JunBuilding Search for AI Agents with Exa CEO Will BrykExa CEO Will Bryk explains how his company is building search infrastructure purpose-built for AI agents rather than human consumers, arguing that Google's two decades of human click data form a competitive moat that becomes largely irrelevant in the agent market. Bryk contends that agents searching for thousands of results at a time, requiring near-perfect reliability and controllable APIs, represent a fundamentally different use case than consumer search.Tue 2 JunAI Agents and the Fight for Customer DataFivetran CEO George Frazier joins the episode to argue that AI agents are triggering a damaging and strategically incoherent wave of API restrictions from SaaS vendors, with SAP announcing a blanket ban on agent access except where specifically approved and Salesforce tightening policies it once kept open. Frazier contends the locked data is context for agents rather than training data, and that almost no companies train their own models, making the lockdowns harmful to customers.Tue 19 MayBen Horowitz on AI Infrastructure, Economics and The New Laws of SoftwareBen Horowitz makes the case that a foundational rule of software development, that you cannot throw money or engineers at a software problem to solve it, has been broken by AI, meaning a product with five to ten years of competitive runway may now have only five weeks before disruption.Tue 12 MayAI Infrastructure, Distribution, and the Next Wave of SoftwareJennifer Li, general partner at Andreessen Horowitz, makes the case that existing storage, compute, and orchestration infrastructure was never built for AI workloads and is being rebuilt in real time, underpinning the firm's $1.7 billion infrastructure allocation within its $15 billion fund. She distinguishes the current cycle from the dot-com era by noting that 90 percent of today's AI infrastructure buildout is pre-committed versus roughly 3 percent during the late 1990s fiber expansion.Tue 5 MayFrom Vector Databases to Knowledge Engines: The Next Layer of AIThe discussion highlights the shift toward knowledge engines in AI, exemplified by Pinecone's evolution from a vector database, which lacks context, to a knowledge engine that enhances efficiency in data retrieval. The introduction of NoQL streamlines agent communication, increasing task completion rates significantly. Additionally, there is an emphasis on the importance of trust and security in AI deployments, with plans for democratizing access and standardizing tools, paving the way for a Cambrian explosion of vertical AI applications.Tue 28 AprWhy We Need Continual LearningThe episode emphasizes the critical need for continual learning in AI, highlighting the differences between human learning and the current static nature of AI models. Malika discusses her research on adaptive learning methods and challenges AI's ability to retain new information post-deployment. Additionally, the conversation touches on in-context learning and its limitations, alongside the significance of developing adaptable AI systems for effective integration in rapidly evolving industries and fostering a growth mindset for personal and professional development.Tue 21 AprThe Agent Era: Building Software Beyond Chat with Box CEO Aaron LevieThe conversation emphasizes the importance of developing software that integrates both agent and human interfaces as AI capabilities evolve. Concerns about job displacement and the necessity for new skills in managing automated processes are prevalent, alongside the need for better frameworks for human-AI interactions. Additionally, discussions around AI and integration highlight the risks of shared systems, emphasizing the importance of reliability and security in software development to harness AI's potential while addressing its current limitations.Wed 8 AprRethinking Git for the Age of Coding Agents with GitHub Cofounder Scott ChaconThe conversation with GitHub cofounder Scott Chacon focuses on rethinking Git's interface for modern coding agents, highlighting the need for improved usability and communication in code reviews. It emphasizes the potential of AI-assisted coding to enhance developer workflows and address gaps in collaboration. Additionally, discussions on evolving Git features reflect the growing trend of automated tools in software development while acknowledging the challenges of managing agent output and inter-team communication.Wed 1 AprHow AI Is Reshaping IT Services from the InsideThe discussion emphasizes how AI is reshaping IT services by enhancing productivity and paving the way for a new model for Managed Service Providers (MSPs) that integrates software, automation, and human expertise. The evolving MSP landscape presents opportunities for growth by addressing operational challenges and moving from reactive to proactive service approaches. Additionally, while the dominance of SaaS continues, the integration of AI and automation is vital for overcoming adoption barriers and optimizing IT and security services.Tue 24 MarPatrick Collison on Stripe’s Early Choices, Smalltalk, and What Comes After CodingPatrick Collison discusses Stripe's early technology choices, emphasizing how foundational programming decisions, such as the shift from Rails to Smalltalk, have lasting impacts on performance and functionality. He also explores the future of coding, advocating for advanced development environments and AI integration that enhances programming efficiency. The conversation touches on the potential of AI, particularly in simplifying code maintenance and improving productivity, while noting challenges in realizing these benefits across industries and technologies like cryptocurrencies and blockchain.Thu 19 MarOpenClaw: Why the Internet Isn't Built for AI AgentsThe discussion emphasizes OpenClaw's potential as an AI agent that enhances user experiences through improved task management and integration capabilities. Security challenges in AI systems and the necessity for new models to facilitate agent interaction are critical points, especially regarding sensitive data and permissions. Additionally, the limitations of existing platforms and the need for startups to innovate in security and access control highlight both opportunities and concerns for the future of AI, cryptocurrencies, and blockchain technology integration.Tue 17 MarWhat's Missing Between LLMs and AGI - Vishal Misra & Martin CasadoThe discussion highlights the limitations of large language models (LLMs) in achieving artificial general intelligence (AGI), emphasizing the need for continual learning and a shift from correlation-based learning to causation. Misra illustrates how LLMs function, using examples like retrieving database queries and Bayesian updating, while stressing their fixed nature post-training. The conversation also explores the theoretical aspects of building causal models to advance understanding beyond mere correlations, crucial for developing AGI and innovative frameworks in machine learning.Tue 10 MarReplit's CEO on Vibe Coding, Wealth Building, and What Most People Get Wrong About AIAmjad Massad, CEO of Replit, emphasizes the potential of AI to democratize coding and empower individuals by streamlining code generation, enabling rapid problem-solving. He advocates for wealth building through user-friendly coding tools and entrepreneurship, asserting that personal ownership and equity are key. The conversation also touches on the decentralization of technology, critiquing traditional career paths while highlighting the transformative role of AI in creating new opportunities and fostering creativity.Tue 3 MarJack Altman & Martin Casado on the Future of VCJack Altman and Martin Casado explore the future of venture capital, emphasizing the shift toward specialization to attract top talent amidst market growth. They discuss the rising importance of media presence for VCs and its impact on investment success, particularly regarding AI. The conversation also addresses the challenges and opportunities presented by AI investments, highlighting the need for strong infrastructure and effective communication channels for portfolio companies.Tue 24 FebAI’s Capital Flywheel: Models, Money, and the Future of PowerThe podcast discusses the intense talent competition within the AI industry, where startups face pressure to achieve rapid growth while securing crucial funding for breakthrough capabilities. It highlights the investment dynamics favoring model companies and the need for a hybrid investment approach in the fragmented market. Lastly, the potential risks of AI oligopolies due to dominant models, alongside the rise of open-source alternatives fostering competition, are examined, emphasizing the future interplay of power between models, money, and market influence.Thu 19 FebDurable Execution and the Infrastructure Powering AI AgentsThe discussion emphasizes three key topics: durable execution as crucial for managing state in long-running AI applications, highlighting its role in enhancing reliability and resilience in microservices; the evolution of AI agents towards specialized, autonomous operators requiring advanced state management and operational insights; and the growing significance of context in AI performance, which fosters better outcomes through effective task management and collaboration in internet-scale transactions.Tue 17 FebEvals, Feedback Loops, and the Engineering That Makes AI WorkThe conversation delves into the engineering essentials of AI, emphasizing evaluation (evals) and feedback loops, crucial for refining models rather than merely employing brute force methods. Ankur Goyal critiques the AI industry's reliance on data and computational power over traditional engineering practices, highlighting concerns around model complexity and integration into enterprises. The discussion also touches on market dynamics and challenges in adopting AI technologies, alongside investor pressures on growth metrics and profitability.Tue 10 FebSam Altman on Sora, Energy, and Building an AI EmpireSam Altman discusses OpenAI's vision for personal AI and the essential infrastructure for developing Artificial General Intelligence (AGI), highlighting Sora's role in this strategy. He emphasizes AI's potential in scientific discovery and the evolution of AI-human interfaces beyond chat. The conversation also addresses the regulatory landscape surrounding AI-generated content and the dynamics of monetization and user engagement, alongside the need for diverse energy sources, particularly as AI technology advances.Tue 3 FebWhy This Isn't the Dot-Com Bubble | Martin Casado on WSJ's BOLD NAMESThe podcast discusses current perceptions of an AI investment bubble compared to the dot-com era, with Martin Casado emphasizing stronger market fundamentals today. He highlights the crucial need for AI infrastructure spending to achieve revenue growth by 2030 and notes significant investments in data center capabilities. Additionally, he reflects on the transformative potential of generative AI, cautioning that while excitement is growing, financial sustainability remains a concern amid increasing debt levels in infrastructure development.Tue 27 JanMartin Casado on the Demand Forces Behind AIThe discussion with Martin Casado emphasizes the accelerating demand for AI and its implications for infrastructure, highlighting the challenges of scaling technology amid resource shortages. A significant shift towards AI agents in decision-making raises concerns about control and visibility. The conversation also notes the evolving role of engineering in software development and the potential need for consumption-based pricing models as AI becomes more prevalent. Regulatory hurdles are seen as a crucial barrier to innovation, particularly in comparing advancements between the U.S. and China.Fri 23 JanHow Mintlify Is Rebuilding Documentation for Coding AgentsThe podcast discusses Mintlify's efforts to enhance software documentation for coding agents and AI, addressing challenges around customization and user needs. Co-founder Han Wang reflects on the evolution of documentation, the significance of self-updating content, and the transition towards broader knowledge management solutions. The conversation also touches on user expectations shaped by AI advancements, as well as insights gained from collaborating with major tech clients like Microsoft and Coinbase.Thu 22 JanInferact: Building the Infrastructure That Runs Modern AIThe discussion highlights the development of VLM, an open-source inference engine aiming to standardize AI infrastructure, emphasizing community collaboration and addressing challenges in scheduling and memory management for large models. Insights into the importance of open-source technology are presented as essential for fostering innovation and standardization in AI. Additionally, the role of InfraRact is outlined in creating a universal inference layer, enabling seamless interactions across diverse hardware while underscoring challenges related to inference at scale.Tue 20 JanHow Should AI Be Regulated? Use vs. DevelopmentThe discussion highlights the pressing need for effective regulation of AI, emphasizing the distinction between regulating its use versus development to prevent stifling innovation. Experts warn that current U.S. regulatory uncertainty is driving talent towards Chinese alternatives, risking competitiveness. Additionally, there's a call for technology-neutral policies that encourage open source innovation while addressing misuse, drawing on historical lessons from cybersecurity and the importance of defining risks associated with AI technologies.Tue 13 JanMichael Truell: How Cursor Builds at the Speed of AIThe discussion with Michael Truell, CEO of Cursor, underlines a pivotal moment in innovation, akin to the "iPod moment," emphasizing Cursor's commitment to sustainable business practices while positioning itself as a leading AI coding provider. Truell highlights the competitive landscape against major tech firms, advocating for practical applications amid rising AI development. Additionally, Cursor's unique hiring strategies and focus on acquiring top talent reflect a shift in growth and startup acquisition models, enhancing their innovation capabilities in the tech sector.Tue 6 JanDylan Patel on the AI Chip Race - NVIDIA, Intel & the US GovernmentThe discussion highlights Nvidia's strategic collaboration with Intel, marked by a significant investment to enhance capabilities in AI chip development and data center products, amidst competitive pressures from AMD and Huawei. Dylan Patel underscores the critical role of market adaptability and customer engagement for Nvidia's success, especially in light of U.S. regulatory challenges impacting foreign chip purchases and the evolving semiconductor landscape. The dialogue also addresses the implications of performance metrics in the AI GPU market, with a focus on delivery efficiency and infrastructure needs.Tue, 30 Dec 2025Feed Drop from The Generalist: Why a16z's Martin Casado believes the AI boom still has years to runMartin Casado from A16Z asserts we are in the early stages of an AI boom, similar to the 1996 dot-com era, stressing the importance of aligning investments with actual AI value creation to capitalize on market potential. He highlights the massive opportunities in AI coding and expresses concerns over Chinese dominance in open-source AI. The discussion also critiques current U.S. policies that may hinder innovation, emphasizing the need for a proactive approach in the AI landscape and recognizing the challenges of successful adoption in technology ventures.Tue, 23 Dec 2025Fei-Fei Li: World Models and the MultiverseThe conversation with Fei-Fei Li highlights the significance of spatial intelligence in AI, advocating for a shift from language-based models to those that prioritize world modeling for applications in robotics and navigation. It underscores the importance of 3D technology in creating immersive environments, essential for effective spatial task performance. Additionally, advancements in 3D computer vision, including Neuradian Fields, are emphasizing the need for interdisciplinary expertise in AI and graphics to tackle complex challenges in the field.Tue, 16 Dec 2025Building the “See Something, Say Something” AI for Every CameraThe episode discusses the transformative potential of AI in physical security, particularly through the development of the "see something, say something" initiative, which aims to enable cameras to detect and report suspicious behaviors. Sheikhar Shrestha's company, Ambient, integrates advanced vision language models (VLMs) to enhance security systems, allowing for proactive monitoring and timely interventions. Additionally, the conversation touches on technological advancements in AI forensics and privacy considerations, while emphasizing the industry's shift towards automation and efficient threat detection in complex environments.Wed, 10 Dec 2025The AI That Found A Bug In The World’s Most Audited CodeThe podcast discusses an AI, Aardvark, that successfully identified a bug in OpenSSH, highlighting significant implications for internet security and open-source software. It outlines the evolution of AI capabilities in vulnerability detection, particularly the transition from GPT-3 to models like GPT-4 and anticipated advancements with GPT-5. Additionally, it emphasizes AI's role in automating cybersecurity tasks and addressing the talent shortage in the field, illustrating a critical shift toward leveraging AI to enhance proactive security measures.Fri, 5 Dec 2025The Death of Data Gatekeeping: AI Makes Everyone An Analyst | Hex CofounderThe discussion emphasizes the transformational role of AI in democratizing data access and analysis, highlighting the shift from traditional BI tools to user-friendly platforms like Hex that enhance collaboration and simplify workflows. Key themes include the need for trustworthy data insights through context management and governance, and the evolution of data teams into "context engineers" to improve decision-making. Additionally, the integration of AI is positioned as critical for reducing costs and improving access in the modern data landscape.Tue, 2 Dec 2025Why Social Engineering Now Works on MachinesIan Webster discusses the growing security challenges posed by AI agents, predicting 2026 as the "year of the agent". He emphasizes the inadequacy of traditional security methods against social engineering tactics, advocating for tools like Promptfoo to aid in securing AI interactions. The conversation highlights the increasing risks of data leakage and vulnerabilities in AI systems, urging developers to adopt dynamic security strategies rather than fixed models. Additionally, it addresses the creative nature of social engineering and its implications for AI security.Fri, 28 Nov 2025“Anyone Can Code Now” - Netlify CEO Talks AI AgentsMatt Billman, CEO of Netlify, highlights the shift in software development due to AI, where traditional coding skills are less vital as tools become accessible for non-developers. He introduces "agent experience" (AX), emphasizing AI agents as key product users, transforming documentation and onboarding. The rise of Large Language Models (LLMs) and autonomous coding reflects an evolving developer paradigm, underlining the necessity for humanistic skills and adaptability in a rapidly changing landscape fueled by AI and blockchain innovations.Tue, 25 Nov 2025From Code Search to AI Agents: Inside Sourcegraph's Transformation with CTO Beyang LiuThe discussion centers on Sourcegraph's evolution from a code search engine to an AI agent, AMP, utilizing large language models to enhance coding efficiency. It highlights the integration of agentic LLMs for improved developer productivity and the critical balance between model performance and user interaction. Additionally, concerns about the reliability of AI models and the implications for market efficiency are addressed, alongside the need for clear regulations in AI and a focus on open-source development to foster innovation.Fri, 21 Nov 2025Ryo Lu (Cursor): AI Turns Designers to DevelopersThe conversation with Ryo Liu from Cursor highlights three key topics: the merging of design and development roles through AI, which facilitates quicker transitions from designs to functional products; the need for unification in fragmented software development processes to enhance collaboration; and the evolving nature of tools like Cursor, which aim to simplify complex design tasks while integrating AI to improve user accessibility and prototype efficiency. These insights emphasize AI's potential impact on enhancing creativity and streamlining workflows in design and development.Tue, 18 Nov 2025How Foundation Models Evolved: A PhD Journey Through AI's Breakthrough EraOmar Khatab explores the evolution of AI foundation models, emphasizing the importance of human-designed systems that prioritize effective communication, retrieval, and training while critiquing the belief that larger models will lead to AGI. He advocates for programmable tools and a structured framework like DSPY to clarify user intent and enhance AI's reasoning capabilities. Khatab also discusses the need for collaboration in AI development and the potential implications for user discernment in the context of evolving model capabilities.