a16z
Showing 1–15 of 1048 transcripts.
- 32 min
Recursive Self-Improvement and the Future of AI
Martin Casado, Dhaval Adjodah, Owen He
This team presents Recursive Self-Improvement as a closed-loop system where the AI optimizes its own cognitive architecture alongside physical hardware and prompts, distinguishing their approach from human-guided autocatalysis. By employing a multi-agent framework with a dedicated "Viceroy of Variance" to stress-test results, the group reports early empirical evidence of simultaneous performance gains and cost reductions that defy traditional diminishing returns. While acknowledging physical limits like Planck's constant and Amdahl's Law, the speakers assert their first-principles mathematical foundation enables domain-specific acceleration, with verified updates expected imminently following rigorous internal replication.
- 53 min
Former Intel CEO: Why This is the Best Time to Build Hardware
Raghu Raghuram, Guido Appenzeller, Pat Gelsinger
Intel CEO Pat Gelsinger warns that the lack of secure power infrastructure will soon trigger a wave of data center project defaults, forcing the industry to adopt 800-volt DC distribution and revive nuclear power to meet the 4% annual capacity growth required for AI. Simultaneously, the sector faces a three-decade memory stagnation and physical manufacturing bottlenecks that are driving a shift toward optical I/O, 3D-integrated memory structures, and a rapid industry consolidation into a few dominant platforms. As AI agents evolve to blur the lines between training and inference, Gelsinger predicts a hardware renaissance focused on solving these thermal and power constraints while replacing copper interfaces with optical switching.
- 56 min
Building the Cloud for AI Agents | AWS CEO Matt Garman
AWS executive leadership outlined a strategic pivot toward nurturing startup ecosystems and agentic workflows by reserving critical GPU capacity and deploying $220 billion in capital expenditures through 2026. The cloud provider introduced new infrastructure components like Tranium chips for inference and simplified onboarding tools while addressing supply chain bottlenecks through direct energy procurement and diversified customer allocation. This approach aims to secure long-term enterprise adoption by prioritizing machine-speed security, data privacy guarantees, and internal "frontier teams" that utilize agent-first software development to accelerate product velocity.
- 48 min
Building Cyber Defense for the Agentic Era
Armaden, founded by former Mandiant CEO Kevin Mandia, deploys autonomous AI agents to execute rapid "hyperattacks" that identify and verify zero-day exploits at Fortune 500 sites before adversaries can leverage them. The company's dual strategy combines offensive red teaming with automated blue team response systems to close security gaps in real time, effectively replacing traditional periodic penetration testing with continuous, human-grade threat simulation. By utilizing agents trained on real-world kill chains, Armaden aims to outpace the accelerating speed of AI-driven nation-state and criminal attacks while maintaining strict operational control to prevent rogue behavior.
- 50 min
The Current State of Consumer AI
Elena Burger, Olivia Moore, Josh Elman
Current AI market dynamics reveal a shift from productivity tools to autonomous agents, driven by high-spending power users who generate a disproportionate share of revenue. While OpenAI leads the field with a $1 billion annual ad run rate and dominant subscription numbers, competitors like Anthropic capture niche enterprise value through premium pricing models. Future growth hinges on overcoming privacy barriers to unlock network effects and transitioning business models from subscriptions toward ads and transaction fees as inference costs decline.
- 48 min
Why Specialized AI Could Beat The God Model
Erik Torenberg, Alex Atallah, Amjad Masad
Following Stripe's acquisition of OpenRouter, the combined entity aims to merge payment infrastructure with AI inference to support enterprises blending proprietary and open-weight models for cost efficiency and data sovereignty. Industry leaders advocate for a shift away from universal "God agents" toward specialized, vertically focused systems that utilize structured output and smaller decision models to mitigate deception risks and ensure psychological accountability. This strategic pivot enables companies to deploy bespoke, low-maintenance tools that reduce vendor lock-in while addressing the growing maturity of internal AI practices.
- 59 min
Why Building an AI Agent Is Easier Than Deploying One
Seema Amble, Elena Burger, Vladimir Keil, Vlad Kyle
Leo is an AI-native startup that deploys a multi-agent architecture to solve the 80% of procurement work ignored by legacy ERPs, moving beyond simple data retrieval to handle complex judgment calls and strategic negotiations. By utilizing a human-in-the-loop adoption strategy, the company has enabled Fortune 500 enterprises to automate both indirect spend and high-stakes direct sourcing while maintaining necessary trust and accuracy. With a business model focused on end-to-end autonomy rather than point solutions, Leo aims to capture the trillion-dollar opportunity of converting unstructured procurement exceptions into measurable financial savings.
- 45 min
Why One Internet Pioneer Thinks the Original Model Broke
Joel de la Garza, Barrett Lyon
Barrett Lyon, a former commercial pilot and cybersecurity entrepreneur, is launching DocsNet to construct an anti-fragile, decentralized mesh network that bypasses traditional internet protocols and centralized surveillance. The system utilizes 26 globally distributed, owner-operated data centers and proprietary AI agents to facilitate direct, ephemeral device-to-device communication with zero-latency file transfers and post-quantum encryption. By eliminating middlemen and carrier-grade NAT, Lyon aims to create a resilient infrastructure that protects free speech and privacy against creeping information totalitarianism and the data-extraction models of the current advertising economy.
- 52 min
AI, Infrastructure, and the Next Investment Cycle
David George, Sarah Wang, Alex Immerman, Santiago Rodriguez
Leading financial analyst Sanjay Chattopadhyay outlines a transformative economic landscape where eight U.S. tech giants drive 90% market growth alongside a projected $90 trillion infrastructure gap extending through 2040. With hyperscaler capital expenditures surging toward $1 trillion annually and private valuations for top AI firms reaching $2.4 trillion, the sector is pivoting from experimental deployments to quantifiable efficiency gains and massive capital backlogs. Consequently, the event positions investors and executives to navigate a bifurcated market where fundamental earnings, not valuation expansion, now underpin S&P 500 performance while signaling delayed public debuts for private giants.
- 50 min
What Would Make an AI Assistant Worth Paying For?
The consumer AI agent market has rapidly expanded to over 120 agents, driven by intense user demand for administrative automation and cost-saving capabilities rather than pure efficiency gains. While current operational costs remain prohibitively high at approximately $20 per user daily, major players like Instinct and Muse are leveraging free pricing models to displace competitors while navigating platform conflicts with tech giants such as Amazon. Looking ahead, the sector is poised to shift from horizontal general assistants to specialized, high-value niche services, ultimately evolving toward a new infrastructure for autonomous agent-to-agent interactions that redefine consumer workflows.
- 42 min
How Jev Turns AI Into Software That Gets Things Done
Jev, Ben Horowitz, Martin Casado, Diogo Almeida
TypeSafe founder Diogo Matos critiques current AI automation for prioritizing code generation over semantic intelligence, arguing that tools like GitHub Copilot fail to solve basic economic tasks despite high benchmark performance. To address this gap, his company introduces Jev, a new software primitive that embeds an intelligent layer directly into code to enable deterministic, intent-driven workflows rather than static output. Matos positions this shift as an "inverse SaaS apocalypse," predicting that pragmatic probabilistic programming will transform industries by automating complex background processes while rejecting the narrative of a singular, general superintelligence.
- 55 min
Why AI’s Next Breakthroughs Could Come from Outside the Big Labs
Erik Torenberg, Aaron Levie, Martin Casado, Steven Sinofsky
The discussion evaluates the precarious intersection of AI regulatory timing, evolving cybersecurity threats, and shifting software architectures, warning that premature rules may stifle innovation while failing to address existential risks. Experts highlight how agent swarms and covert channels necessitate a "secure by design" operating model, yet argue that historical precedents suggest policy often arrives only after catastrophic failure. Consequently, the industry faces a complex political landscape where vague terminology and ambiguous safety stances risk regulatory capture before a cohesive national narrative on artificial intelligence can emerge.
- 50 min
How to Spot Exceptional Talent Before Everyone Else
Erik Torenberg, Josh Elman, Olivia Moore, David Booth
Cosign is a new digital platform designed to formalize professional reputation by replacing binary social connections with granular, peer-to-peer endorsements that validate career impact, crisis-level trust, and future potential. Leveraging AI to aggregate public data and a strictly positive feed, the system constructs durable professional profiles that allow early endorsers to build social capital as their mentees succeed. Unlike existing networks, Cosign targets the tech and startup ecosystem with a "living directory" model that prioritizes relationship depth over volume, aiming to make talent discovery a permanent and searchable asset.
- 47 min
Replit CEO Amjad Masad on What Young People Should Learn in the AI Era
Amjad Masad, Erik Torenberg, Gagan Biyani
This presentation critiques traditional higher education for suppressing deep inquiry and advocates for "The Academy," a competency-based alternative where 18-to-22-year-olds design autonomous, project-driven curricula. Leveraging AI to enable personalized learning and self-driving organizational models, the proposed framework replaces prescriptive grading with a two-year exploration window focused on virtue ethics and character cultivation. Ultimately, the speaker argues that institutions must shift from enforcing conformity to identifying and nurturing individuals with deep, persistent focus and strong interpersonal skills to address a societal reset.
- 43 min
Why Education Has to Change
Ben Horowitz, Erik Torenberg, Gagan Biyani
Horowitz and Jason Academy is launching a for-profit educational entity in the San Francisco Bay Area under CEO Gagan Biyani to train Gen Z builders through an immersive, output-driven curriculum aligned with AI era demands. The program partners with industry leaders like Nvidia and Stripe to integrate real-world problem-solving and leadership development, replacing traditional lectures with a "pull model" that prioritizes tangible creation over memorization. By operating as a market-aligned entity, the Academy aims to produce high-impact graduates capable of launching companies or securing elite roles, ultimately serving as a scalable blueprint for future educational innovation.