The a16z Show

David George × Will Gaybrick

Stripe’s first-half signups grew 50% YoY, while the median 2026 cohort generates 50% more revenue than 2025’s and its new SaaS-platform cohort is 103% larger.Stripe Minions produce 7,000 weekly one-shot PRs, KAI lifted seller productivity 20%, and Machine Payments Protocol, Link Agent Wallet, and Tempo offer potential catalysts as agentic commerce adoption and execution remain unproven.

Ben Horowitz × Travis Kalanick × Erik Torenberg

Adams is positioning “industrial AI” as an atom-based computer, with robotic food production at roughly $6–8 per meal targeting delivered meals near grocery-store cost and mining automation pitched as 20% more gold per year.Transport becomes the platform’s wheelbase across multiple hundred-billion-dollar industries, but the thesis still hinges on proving autonomy faster than Waymo and overcoming resistance to change.

Elena Burger × Andy McCall × Joe Schmidt

Joe Schmidt’s 2x2 contrasts lighthouse AI sales, where proof travels amid buyer exposure, with land grabs, where budgets make the math decisive.Andy McCall’s Samsara example links 2016–2019 ELD mandate and mid-market focus to faster feedback despite AT&T and Verizon.AI boards and buy-by-X deadlines may open a software window, but founders need ACV discipline and hard-dated POCs with upfront criteria.

Anish Acharya × Garry Tan

Pure per-seat SaaS may disappear within five or 10 years; Tan says skill files could help two or three people reach $15M ARR in about four months, if defensibility comes from data or network effects.His token-maxing model costs $50,000–$100,000 a year and points to 2027’s “harness wars,” but 20-year diffusion and organizational bottlenecks remain the timing risk.

Angela Strange × Gabriel Vasquez × Alejandro Maza Ayala

Kavak now instantiates 100 to 200,000 customer-specific agents daily; 96% of interactions and 95% of transactions are fully agent-handled.These sales agents convert 2.1× better than humans, triple NPS and satisfaction, and approve car loans in under three minutes.Yet evals consume roughly equal resources to agents, and Opus 4.5 made Maza scrap two years of architecture, highlighting redesign’s upside and obsolescence risk.

Joel De La Garza × Dylan Ayrey × Feross Aboukhadijeh

Frontier models are collapsing the expertise barrier to cyberattacks, with exposed credentials and software supply chains offering cheaper routes than zero-days.AI is compressing vulnerability discovery into same-day exploitation while npm worms can recruit developer agents and propagate through packages; 2027 human 2FA, registry funding, and blue-team access to comparable tools are key defenses to monitor.

Elena Burger × Matt Bornstein × Simon Mo

vLLM has become the execution layer linking more than 1,000 open-weight model architectures to GPUs from NVIDIA, AMD, Google, Amazon, Intel, and others, making inference a strategic systems layer.Open weights increasingly offer controllable latency, data, security, and fine-tuning rather than merely cheaper tokens, while licensing and moderation pressures test whether the ecosystem can fund frontier development and trusted specialized use cases.

Sarah Wang × Kimberly Tan × Jesse Zhang × Ashwin Sreenivas

Decagon runs 90% of production workflows on open-source models, where task-specific fine-tuning delivers higher accuracy, lower latency, and lower cost than frontier systems on bounded jobs.Frontier models remain the discovery engine, while Decagon’s continuously rebuilt model factory and enterprise infrastructure turn deployment pain into reusable product; migration will remain constrained by proprietary data, security, and governance.

Alex Rampell × Olivia Moore × Steijn Pelle × Frédéric Renken

Lassie monetizes an understaffed labor budget: U.S. dental practices spend roughly $200,000 annually on administration, while its agent already sells for five figures.Its reported 98% automation and read-write workflow integrations could expand from dentistry, but distribution, onboarding, proprietary knowledge, and paper payments remain key execution risks.

Martin Casado × Fei-Fei Li × Yunzhu Li

World Labs is extending its spatial-intelligence stack into robotics by bringing SpAItial inside rather than manufacturing robots.The combination pairs Marble’s geometrically consistent worlds with real-to-sim-to-real robotics expertise to address scarce data and slow, costly evaluation.Near-term traction depends on proving aligned simulation in structured factories, warehouses, hotels, and restaurants, while homes and human-level efficiency remain distant risks.