Fully Connected 2026 | Day 2 Full Replay
Summary
The keynote explores the concept of 'ambition' in technology, detailing the massive infrastructure shifts required to support advanced AI and agentic workflows. Speakers emphasized that the industry is moving toward a highly complex, interconnected stack where compute, models, and data must be co-designed. Key technical insights include the prediction that web traffic will be dominated by AI agents, necessitating a fundamental redesign of web search at a 'thousandx' scale, and a strong warning regarding the 'wild west' state of AI security across multi-tenant cloud environments.
Key takeaways
-
The Shift from Pull to Push Web Interaction
16:50
The web is predicted to move from a 'pull' model (where users actively search for information) to a 'push' model, where proactive AI agents receive actionable information directly (10:10).
-
AI Agents Require Dedicated Web Search Infrastructure
16:50
Building a web search for agents requires a GPU-native index operating at massive scale, fundamentally different from traditional search systems (10:10).
-
AI Infrastructure Requires Holistic Co-Design
16:50
The most complex systems require understanding and optimizing the entire stack—from silicon and kernels to the cloud layer and application deployment—rather than treating layers in isolation (10:10).
-
Security is a 'Wild West' Problem
16:50
The rapid evolution of AI has created significant security vulnerabilities across multi-tenant clouds. Testing revealed that numerous clouds could allow unauthorized access to storage systems and job names, highlighting the need for a 'top bottom security view' (10:10).
-
The Importance of 'Options' in Architecture
16:50
When designing systems, the most valuable aspect is maintaining flexibility and understanding the 'options' or alternative use cases for underlying resources (e.g., GPUs), rather than optimizing for a single use case (10:10).
Technical details
-
Compute Infrastructure
728s
CoreWeave built its cloud infrastructure initially using GPUs, pivoting from crypto mining to general compute services to support demanding applications (7:28).
-
AI Workload Management
728s
CoreWeave offers tooling like Mission Control and Arena to manage workloads across the entire AI lifecycle, from training through inference to agents (7:28).
-
Agentic Web Search
1010s
Parallel Web Systems is building a 'Google for agents,' focusing on a GPU-native index to handle web search requests at a 'thousandx' scale (10:10).
-
Cloud Security Vulnerabilities
1010s
SemiAnalysis tested over 80 clouds and found that multiple providers could be vulnerable to unauthorized access to storage systems and job names, regardless of the cloud's supposed security status (10:10).
Mentioned resources
- CoreWeave Forge
- CoreWeave Builder Resource Center
- SemiAnalysis
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.