Latest Interviews
Showing 1–5 of 5 transcripts.
Clear all filters- Sequoia Capital51 min
Google DeepMind's Logan Kilpatrick: Why the Model Eats the Harness
Google has launched its "agentic era" centered on the Anti-Gravity harness and the new Omni unified model, enabling autonomous actions across its ecosystem rather than simple API interactions. Key shifts include the introduction of the 3.5 Flash coding model, the transition from maximizing user time to optimizing task completion, and the deployment of a single system for diverse generative media tasks. Led by leadership emphasizing scientific rigor under Demis Hassabis and Sundar Pichai, the initiative aims to replace specialized legacy tools with native, world-understanding agents that augment rather than cannibalize human activity.
- Sequoia Capital46 min
How Cursor Trained Composer on Fireworks: Distributed Infrastructure for High-Performance RL
Federico Cassano, Dmytro Dzhulgakov, Sonya Huang
Cursor has pivoted from a pure application company to a foundation model developer by training Composer 2, a specialized software engineering model built on a Kimi 2.5 base using a global, asynchronous pipeline to maximize compute efficiency. The team deployed a custom infrastructure that integrates Reinforcement Learning from real-time user feedback and simulated rollouts while overcoming synchronization challenges through lossless delta compression and custom GPU kernels. This strategic approach allows Cursor to saturate model capacity with high-value coding data, achieving competitive performance at a fraction of the cost of general-purpose models while moving the industry toward baked-in specialized behaviors rather than prompt engineering.
- Sequoia Capital49 min
The Breakthroughs Needed for AGI Have Already Been Made: OpenAI Former Research Head Bob McGrew
Bob McGrew, Stephanie Zhan, Sonya Huang
Bob McGrew defines 2025 as the "year of reasoning," a pivotal shift where immediate compute utilization and tool-augmented chain-of-thought capabilities drive rapid progress toward AGI while pre-training faces diminishing returns. This paradigm reframes post-training as a behavioral challenge and positions AI agents to democratize intelligence by pricing services at compute costs rather than professional market rates. Consequently, economic value accrues to application layers requiring proprietary enterprise context, while sectors ranging from robotics to software engineering transition to hybrid human-agent workflows that leverage these new efficiency gains.
- Sequoia Capital1h 0m
AI, Security and the New World Order ft. Palo Alto Networks’s Nikesh Arora
Nikesh Arora, Sonya Huang, Pat Grady, Jim Goetz
Nikesh Arora outlines a shifting AI landscape where falling development costs enable specialized models while warning that premature agency requires rigorous "AI firewalls" to mitigate real-time cyber threats and hallucinations. He details Palo Alto Networks' strategy of acquiring only category leaders under strict co-authorship agreements, a tactic designed to preserve agility amidst a predicted five-year battle between autonomous agents. Additionally, Arora forecasts a regulatory bifurcation for critical infrastructure alongside a market split between enterprise-grade specialized systems and consumer-focused general-purpose AI.
- Sequoia Capital59 min
Factory’s Matan Grinberg and Eno Reyes Unleash the Droids on Software Development | Training Data
Matan Grinberg, Eno Reyes, Sonya Huang, Pat Grady
Factory deploys autonomous software engineering "droids" that leverage existing foundation models to automate unenjoyable enterprise tasks like code review and testing, delivering a 22% increase in engineering cycle speed. The company recently achieved a 19% pass rate on the SWE-Bench benchmark by prioritizing task-specific cognitive architectures over training new models, effectively surpassing previous state-of-the-art performance. Founders Matan Grimberg and Eno Reis position the platform to shift engineering roles toward orchestration, focusing on measurable organizational metrics rather than individual developer replacement.