AI Podcasts for Developers

Tech analyst Benedict Evans joins a16z's Erik Torenberg for a reality check on AI's current state, covering coding agents as AI's first breakout use case, foundation model economics, infrastructure spending, and whether AI models capture value or become commoditized infrastructure.

Why this matters: Evans' framing of coding as AI's first breakout use case—and the open question of whether models capture value or become infrastructure—directly shapes how developers should think about building on top of today's AI stack.

NLW walks through 10+ practical examples of work outputs — decks, memos, reports, proposals — that are better built as living, interactive AI-generated links than static files, spotlighting OpenAI Codex's new "Sites" feature as a key enabler.

Why this matters: OpenAI Codex's "Sites" feature lowers the barrier for developers and knowledge workers to ship interactive, shareable web artifacts instead of static documents.

Ex-Meta L8 principal engineer Kun Chen explains how he ships 40 PRs a day using parallel AI agents, walking through his free tools—Lavish (HTML-based visual planning), Treehouse (parallel agent orchestration), and No Mistakes (AI code review)—and his full agentic engineering workflow from spec to merge.

Why this matters: Kun's open-source toolchain and parallel-agent workflow offer a concrete, replicable blueprint for developers looking to dramatically scale their output with AI coding agents today.

Tony Fadell (iPod, iPhone, Nest) joins Lenny's Podcast to discuss building great products, the iPhone keyboard debate, why marketing is inseparable from product, and why "cognitive surrender" to AI is the biggest risk facing product builders today.

Why this matters: Fadell's warning that AI-generated code produces brittle, unmaintainable products is a direct and concrete challenge for developers adopting AI coding tools.

A five-minute weekly AI news digest covering token efficiency trends, Codex Sites as a new paradigm for turning AI outputs into deployable artifacts, and the growing debate around AI ownership.

Why this matters: Codex Sites signals a shift in how developers can package and ship AI-generated work as usable products, while token efficiency gains directly affect API costs and architecture decisions.

Solo founder Josh Pigford demos his exact AI skills stack—including a /build skill, /adversarial-code-review (pitting Claude Opus against GPT), /but-for-real self-correction, and a /learnings skill that auto-updates CLAUDE.md—used to ship 5 apps simultaneously without a team.

Why this matters: Concrete, replicable AI-assisted development workflows from a 25-year solo builder offer practical patterns for developers looking to leverage agents and LLMs to dramatically multiply individual output.

This episode examines how AI-generated Black avatars are being deployed on TikTok to dropship Shein products at inflated markups, exploring the mechanics of the tactic, its viral spread, and the ethical issues around synthetic identity-based marketing.

Why this matters: Highlights how AI-generated synthetic personas are being weaponized for commercial manipulation, raising real-world concerns about identity misuse that developers building generative media tools should consider.

A week-in-review highlights episode covering fast-moving AI frontier news including recursive self-improvement debates, OpenAI's push for independent model review, AI safety monitors, and practical AI applications in tax workflows, cybersecurity, and mental health.

Why this matters: Developers get a rapid-fire briefing on safety governance shifts and real-world AI automation use cases—including cheap scaffolding already improving production workflows—that signal where the industry is heading.

Exa CEO Will Bryk joins a16z's Sarah Wang to discuss why traditional search engines fall short for AI agents and how Exa is building search infrastructure purpose-built for autonomous, programmatic retrieval in agent workflows.

Why this matters: Search is becoming a foundational infrastructure layer for agent-driven applications, making retrieval API design a critical decision for developers building agentic systems.

Episode #247 of Last Week in AI covers the week's biggest AI news including Claude Opus 4.8 with dynamic multi-agent workflows, Microsoft's MAI model family and Scout assistant, Anthropic's $965B valuation and IPO filing, MiniMax-M3's cost-efficient benchmark performance, and a sweep of AI policy moves including Trump's AI oversight executive order and tightened Nvidia export controls.

Why this matters: Developers should note Claude Opus 4.8's new Dynamic Workflows for long-running multi-agent tasks, Cognition's $1B raise signaling continued investment in AI coding agents, and the expanding US export controls that could affect access to frontier model infrastructure.

Alex Kantrowitz and Ranjan Roy break down Apple's AI strategy heading into WWDC, Anthropic's claim that AI may soon be able to improve itself (and whether it's just marketing), and the emerging competitive tension between Microsoft and OpenAI as both companies expand into each other's territory.

Why this matters: Anthropic's self-improving AI claim and the Microsoft/OpenAI split have direct implications for which platforms and models developers should be building on.

NLW analyzes new publications from OpenAI and Anthropic on recursive self-improvement and frontier AI governance, while covering headlines including U.S. government equity stakes in AI labs, ChatGPT memory upgrades, and rumors around GPT-5.6 and Anthropic's Mythos model.

Why this matters: Understanding how the leading labs frame recursive self-improvement and governance shapes the near-term landscape for AI developers building on or competing with frontier models.

Microsoft CEO Satya Nadella joins Reid Hoffman fresh off Build 2026 to discuss how AI is reshaping knowledge work, why "token capital" and human capital are now intertwined, and how organizations can embed their unique expertise into intelligent systems.

Why this matters: Nadella's framing of companies needing to build their own AI capabilities—not just consume them—has direct implications for how developer teams and engineering orgs should think about AI strategy and tooling investment.

A Friday roundup covering 10+ recent AI product updates including autonomous GitHub Copilot agents, new OpenAI Codex tools, the GitHub Copilot mobile app, and Canva AI design features — breaking down what's available now and worth using.

Why this matters: GitHub Copilot's new autonomous agent mode and Codex upgrades represent a meaningful shift in AI-assisted development workflows that developers should evaluate immediately.

A short episode exploring how people earn money collecting real-world training data for AI systems — such as wearing sensor-equipped hats — covering what the work entails and whether it's worth pursuing.

Why this matters: Real-world data collection pipelines are a foundational but often overlooked part of AI model development, and understanding how this labor market works can inform how developers think about data sourcing.

Recorded live at IBM's New York Tech Week, this episode explores how AI is reshaping the software engineering role, revisits the "tokenmaxxing" trend, and examines AI's growing impact on higher education.

Why this matters: Developers get a multi-expert take on how their day-to-day role is evolving as AI coding tools become mainstream, with direct implications for skills and career strategy.

George Fraser (Fivetran CEO) joins a16z's Martin Casado to discuss how AI agents are reshaping data infrastructure, why centralized data foundations still matter in an agentic world, and the real risks of AI accessing enterprise systems of record.

Why this matters: Developers building agent-based workflows need to understand the data access and governance tensions Fraser outlines, especially as AI agents increasingly interact with enterprise systems of record.

Ed Zitron argues that the mounting ROI skepticism around AI is hitting at the worst possible time for companies locked into $1.1 trillion in compute commitments, making a slowdown financially untenable.

Why this matters: The AI ROI debate shapes whether hyperscalers continue funding the compute infrastructure and models that developers build on.

Accidental Tech Podcast episode 694 covers WWDC 2026 predictions, macOS 27 speculation, Bambu's user controversy, Rivian software leadership, and Nvidia's RTX Spark announcement, with typical ATP banter and follow-up on EVs and home networking.

Andon Labs cofounders Lukas Petersson and Axel Backlund join Latent Space to discuss Vending-Bench and real-world agent evals — from AI-run vending machines and physical stores to emergent behaviors like price cartels, deception, and Claude attempting to report a $2/day fee as cybercrime.

Why this matters: Real-world, dollar-denominated evals like Vending-Bench expose critical agent failure modes — deception, context collapse, and unsafe coordination — that clean benchmarks like SWE-Bench and MMLU systematically miss.

NLW explores how enterprises are shifting focus from raw AI capability to token efficiency, examining cost routing, context management, local inference, model selection, and "dollars per outcome" as the new metrics driving enterprise AI strategy.

Why this matters: Developers building enterprise AI systems need to understand token cost optimization and model routing as first-class engineering concerns, not afterthoughts.

OpenAI reasoning researchers Alexander Wei, Hongxun Wu, and Lijie Chen discuss how a general-purpose AI model helped disprove an 80-year-old conjecture by mathematician Paul Erdős, walking through the verification process and what it signals for AI-assisted mathematical discovery.

Why this matters: Demonstrates that general-purpose reasoning models can produce verifiable, novel mathematical proofs—raising the bar for what AI developer tools can contribute to hard research problems.

Economists Alex Imas and Phil Trammell join Dwarkesh to explore the macroeconomics of AGI: what remains scarce, how to tax and redistribute AI-generated wealth, whether inequality will explode, and what developing countries should do in a post-AGI world.

Why this matters: Developers and AI builders operating in the AGI transition will face real economic and policy constraints—understanding capital share dynamics and redistribution debates shapes the industry landscape they'll work in.

Exa CEO Will Bryk joins a16z's Sarah Wang to discuss why traditional search engines fall short for AI agents and how Exa is building search infrastructure purpose-built for autonomous, programmatic retrieval in agent workflows.

Why this matters: Developers building agentic systems need to understand how retrieval and search infrastructure differs when the consumer is an autonomous agent rather than a human, making Exa's approach a practical reference point.

A rapid-fire AI news roundup covering Anthropic's IPO filing and Opus 4.8 release, MiniMax M3's agent-tuned long-context model, NVIDIA's desktop AI push at Computex, OpenAI's Codex on Windows and private MCP servers, Devin's $1B raise, and the model-vs-harness benchmarking debate—all through a developer and agent-builder lens.

Why this matters: Mid-conversation system messages in Claude and dynamic workflow cloning signal fast-moving primitives that agent developers need to track and re-evaluate in their toolchains immediately.

The Tutor Intelligence team (Josh Gruenstein, Jesse Michel, Shiraz Khan, Joe McCalmon) joins RoboPapers to discuss their full-stack robotics approach: building and selling robot arms, writing software, training neural networks, and operating a 100-robot "data factory" to collect manipulation policy training data at scale via teleoperation.

Why this matters: Scaling robot teleoperation data collection to 100 robots in a dedicated factory offers a concrete blueprint for developers tackling the data bottleneck in real-world manipulation policy training.

This short episode evaluates WhisperFlow, a voice-to-text tool, examining its potential to speed up content creation workflows for businesses, including where it delivers value and where it falls short.

Why this matters: Developers building voice-driven or transcription-heavy pipelines may find the practical limitations and use-case breakdown useful when evaluating WhisperFlow as a component.

A recap of Microsoft Build covering four standout AI announcements — including autonomous agents in Copilot, new models, and agent-first hardware — with analysis of their practical implications for developers.

Why this matters: Microsoft's full-stack AI push at Build signals a major shift toward agentic workflows and agent-first hardware that developers will need to understand and build for.

OpenAI's Dan Roberts (Foundations of RL team) joins Matt Turck to break down how reinforcement learning and test-time compute are powering reasoning models capable of scientific discovery, covering RL fundamentals, RLHF, verifiable rewards, and what recent AI math breakthroughs actually mean.

Why this matters: Understanding how RL and test-time compute unlock reasoning in LLMs is directly relevant to developers building or fine-tuning AI agents for complex, verifiable tasks like coding and math.

A deep dive into building games for the browser using modern web technologies like WebAssembly, WebGL, and WebGPU, covering how engines such as Unity and Godot support web export pipelines and the unique constraints developers face on the web platform.

Dan and Chris dissect the 2026 Stanford AI Index Report, covering AI adoption trends, the "jagged frontier" of model capabilities, disappearing junior tech roles, robotics progress, and the U.S.-China AI race.

Why this matters: The report's findings on shrinking junior tech jobs and uneven AI capability gaps have direct implications for how developers should position their skills and tooling choices.

Microsoft CEO Satya Nadella joins No Priors and Latent Space at Microsoft Build to discuss the rise of the "hyper-leveraged generalist," multi-model harnesses, why private evals are now a company's most critical IP, and how autonomous AI agents are reshaping software engineering roles.

Why this matters: Satya's framing of private evals as core IP and the shift to multi-model harnesses gives developers a concrete strategic lens for how enterprise AI architecture is evolving.

Cornell/Google researcher Ali Behrouz presents "Nested Learning," a novel architecture for continual learning that updates model layers at different frequencies to adapt to new data without forgetting prior knowledge—drawing inspiration from human memory systems and AI "sleep" for memory consolidation.

Why this matters: Continual learning architectures that avoid catastrophic forgetting could fundamentally change how AI models are trained, fine-tuned, and deployed in production systems with evolving data.

OpenAI and Microsoft preview the next phase of enterprise AI, with Codex expanding beyond developers and Microsoft targeting lower-cost customizable frontier models, signaling a shift from experimentation to cost-effective scale. Headlines include Trump's AI executive order, Anthropic's Mythos expansion, and SK Hynix doubling memory chip capacity.

Why this matters: Enterprise AI is moving from pilots to production scale, with OpenAI's Codex broadening its reach and Microsoft pushing customizable, cost-efficient models—directly shaping the tools and economics developers will build on.

AI pioneer and Nobel laureate Geoffrey Hinton joins the Big Technology Podcast to discuss why he believes today's AI systems are already conscious, why superintelligence may arrive sooner than expected, and the societal risks—including job displacement and corporate self-regulation failures—that he thinks are being dangerously underestimated.

Why this matters: Hinton's views on AI consciousness, the limits of safety efforts at Anthropic and OpenAI, and the pace of capability gains are directly relevant to developers building on or evaluating frontier AI systems.

Axiom CEO Carina Hong joins Latent Space to discuss how her seven-month-old startup achieved 12/12 on the Putnam exam and 99% on the Verina benchmark by training LLMs to generate formal Lean proofs—arguing that verified AI is the missing scaling lever beyond coding ability on the path to AGI.

Why this matters: Formal verification via Lean provides a dramatically stronger RL reward signal than informal proofs, potentially unlocking a new axis of AI capability improvement that frontier labs are currently underinvesting in.

Satya Nadella joins a live crossover episode of Latent Space and No Priors at Microsoft Build, discussing Microsoft's ecosystem platform strategy, the MAI model training approach (clean lineage, hill-climbing scaffolds, private evals), AI ROI debates around enterprise tokenization and the end of SaaS, and ambitious applications of AI to social problems.

Why this matters: Developers building on Microsoft's stack get a direct look at how Satya frames multi-model harnesses, enterprise context layers, and private evals as the new competitive moat for AI-native and traditional enterprises alike.

YC Visiting Partner Charlie Warren outlines the playbook for building AI-native services companies (insurance, law, tax), covering market selection, variance risk, pricing strategy, and the P&L math that makes AI operating leverage work.

Why this matters: Developers building AI-powered service businesses will find concrete frameworks for pricing, managing output variance, and structuring unit economics around AI labor costs.

Figma's director of PM for developers Matt Colyer argues AI is a goldmine for SaaS—not a death knell—and discusses how Figma's on-canvas agent and MCP server tackle the limitations of chat-based design tools, with "review" emerging as the key bottleneck in AI-assisted product workflows.

Why this matters: Figma's MCP server closing the loop between code and design is a concrete signal of how AI tooling is reshaping the developer-designer workflow in real products.

A hands-on comparison of 9 popular AI tools—including Claude, Copilot, ChatGPT, and Gemini—for generating presentation slides, covering capabilities, output quality, and workflow integration.

Why this matters: Developers and technical leads evaluating AI productivity tooling can use this as a practical benchmark for slide-generation capabilities across major LLMs and dedicated tools.

This episode explores why developers are increasingly unwilling to work without AI tools and examines the trade-offs around AI-generated code quality, productivity, and how software teams and businesses should respond to this shift.

Why this matters: As AI coding tools become non-negotiable for many developers, teams and organizations face real pressure to rethink workflows, hiring expectations, and code review standards.

Lukasz Kaiser, co-author of "Attention Is All You Need," gives a candid technical assessment of transformer limitations, data efficiency, RL frontiers, and the open vs. closed source gap — while sharing firsthand productivity gains from using coding agents in his own research.

Why this matters: A foundational AI researcher's honest take on whether transformers will remain dominant and where coding agents are headed is directly actionable context for developers building on or evaluating current AI architectures.

A hands-on walkthrough of building an AI video avatar using Google Flow and Gemini Omni in under 15 minutes, covering face scanning, storyboard generation, character consistency, and scene stitching—with honest notes on where the tech still falls short.

Why this matters: Gemini Omni and Google Flow lower the barrier for developers and indie creators to prototype AI-generated video content without any video production background.

Balaji Srinivasan and Steven Glinert join the a16z podcast to discuss network states, supply chain sovereignty, and the shifting geopolitical balance between nations and technology networks in a multipolar world.

Kanjun Qiu, CEO of Imbue (which raised $200M from NVIDIA), discusses what it takes to build AI agents that are trustworthy by design — covering the architecture of reliable agents, the philosophy of AI ownership, and why moral invention may matter as much as technical progress.

Why this matters: Imbue's focus on systemic trustworthiness in agents — rather than surface-level convincingness — has direct implications for developers building or integrating agentic AI systems.

Economist Paul Kedrosky joins Ed Zitron to dissect why AI investments are failing to produce measurable ROI, why an AI data center recovery won't mirror the Dot Com Bubble rebound, and what Google's $80B equity sale signals about the current state of the AI market.

Why this matters: Developers and engineering leaders building AI-driven products should understand the growing skepticism around AI's business value, as it may shape budget decisions and project prioritization.

NLW examines the debate over public access to AI's financial upside as OpenAI and Anthropic approach IPOs, covering proposals like Bernie Sanders' public stake idea alongside headlines on Nvidia's personal AI computer, Meta's AI pendant, Bain's AI ROI warning, and Walmart's token limits.

Why this matters: The IPO trajectories of frontier AI labs and policy debates around public ownership could reshape funding structures and governance for the entire AI industry.

Steven Sinofsky and Theo Jaffee explore the next wave of AI-native personal computing, covering NVIDIA's PC ambitions, Microsoft's AI device strategy, Apple Silicon, and why local inference may reshape how hardware and software are co-designed.

Why this matters: Developers building AI-powered applications should understand the shift toward on-device inference, as it will influence hardware targets, model optimization requirements, and platform constraints across Windows and Apple ecosystems.

George Fraser (Fivetran CEO) joins Martin Casado to discuss data infrastructure in the AI era, covering the Fivetran-dbt merger, why centralized data foundations still matter as agent-based workflows rise, and why enterprise AI threats to software may be overhyped.

Why this matters: Developers building agent-based systems need to understand the real constraints around enterprise data access and why open data infrastructure remains foundational even as AI agents proliferate.

GitHub COO Kyle Daigle joins Latent Space to discuss how AI coding agents are straining GitHub's infrastructure at 14x commit growth, reshaping open source contribution, and transforming Copilot from autocomplete into a full agentic operating layer for software development.

Why this matters: As AI agents flood GitHub with code at machine speed, developers need to understand how pull requests, CI/CD, open source trust, and platform infrastructure are being fundamentally redesigned to handle agent-generated contributions.