Fully Connected 2026 Day 2: Mike Intrator on Hot Ones, Parag Agrawal, Dylan Patel, and Jenson Button
Summary
The discussion centered on the infrastructure required to support the next wave of AI applications, emphasizing the shift from traditional compute to agent-driven, highly scaled systems. CoreWeave detailed its journey from crypto mining to building a specialized cloud infrastructure, focusing on 'fearless risk-taking' and 'superior risk mitigation.' Key insights included the prediction that web traffic will be dominated by AI agents, necessitating a 'Google for agents' built on GPU-native indexing. Speakers stressed that the industry faces significant fragmentation and security risks in the current AI ecosystem, requiring a holistic, full-stack approach to deployment and scaling.
Key takeaways
-
The Web is Shifting from Pull to Push for Agents
The future of web usage is predicted to move from traditional 'pull' (search queries) to 'push,' where systems proactively deliver actionable information to agents, minimizing the need for agents to search the web themselves. This requires a fundamental shift in web architecture.
-
AI Agents Require Specialized Web Search Infrastructure
22:00
To support AI agents, a specialized 'Google for agents' is needed. This requires building a GPU-native index capable of operating at massive scale, fundamentally different from traditional web search systems.
-
The Importance of Full-Stack Infrastructure Co-optimization
25:40
Successful AI deployment requires viewing the entire stack—from models and kernels to silicon and cloud infrastructure—as a single, interconnected system. Optimization must occur across all layers, not just the model or the compute layer.
-
Security is a 'Wild West' in AI
28:40
The rapid deployment of AI creates a complex security landscape. Concerns include unauthorized access between tenants on shared cloud infrastructure and the need for a top-to-bottom security view that addresses both malicious actors and rogue AI.
Technical details
-
Compute Infrastructure & Scaling
120s
CoreWeave initially leveraged GPUs from crypto mining to fund the development of a cloud platform, allowing them to scale compute for demanding applications like simulation. They provide tooling (Mission Control, Arena) to manage and prove performance for workloads ranging from training through inference to agents.
-
Agent Architecture and Web Indexing
1320s
The core technical challenge for AI agents is accessing and processing web data at scale. Parallel Web Systems is building a GPU-native index to enable web search for agents, allowing them to function as sophisticated information requesters.
-
AI Deployment Challenges and Fragmentation
1420s
The industry is highly fragmented across the model layer, API usage, and application scaffolding. Building robust AI systems requires careful management of these layers, as new models can break existing scaffolding, necessitating deep expertise in co-designing the entire solution.
-
Cloud Security and Multi-Tenancy Risks
1720s
The current cloud environment is described as a 'wild west' regarding security. Vulnerabilities can be exploited across the stack, making it critical for enterprises to assess not only the service provider but also the underlying infrastructure and other tenants sharing the cloud.
Mentioned resources
Channel & topics
Watch on YouTube · Back to latest
This independent, AI-assisted summary is provided for commentary and informational purposes. It may contain errors or omit important context. Please watch the original video for the creator's complete presentation. Video, thumbnail, and related copyrights belong to their respective owners.