# CoreWeave Fully Connected 2026: Everything Announced in 10 Minutes

## Executive summary

CoreWeave announced a comprehensive platform, CoreWeave Forge, designed to manage the entire AI development lifecycle—the 'AI loop' (Run, observe, curate, improve, evaluate). Key announcements include the launch of CoreWeave Forge, the production readiness of the NVIDIA Vera Rubin NVL72, and new tools like CoreWeave Agent Lens and Model Distillation, which aim to improve agent reliability and streamline MLOps for AI applications.

## Key takeaways

- CoreWeave Forge Launch: CoreWeave Forge centralizes the entire AI loop (Run, observe, curate, improve, evaluate) into one connected environment, allowing users to build with chosen models, frameworks, and clouds. It is available for self-service sign-up and starts with a 30-day Pro trial.
- NVIDIA Vera Rubin NVL72 Production Readiness: The Vera Rubin NVL72 is announced as ready for production, with Cognition serving as the first customer. Performance comparisons showed a 4.8x improvement in throughput for inference and a 3.8x improvement in training workloads compared to previous generations.
- CoreWeave Agent Lens and Model Distillation: CoreWeave Agent Lens helps understand agent behavior in production by grouping related failures into patterns. CoreWeave Model Distillation allows teams to use curated production traces to create smaller, specialized models.
- RL Rollouts for Reinforcement Learning: RL Rollouts addresses the slow learning cycle in reinforcement learning by allowing weights to be loaded directly into a running deployment when a new checkpoint is ready, minimizing waiting time for serving.

## Technical details

- AI Loop Process: The core operational cycle is defined as: Run, observe, curate, improve, evaluate, and repeat. CoreWeave Forge integrates these steps into a single platform.
- Performance Metrics (Vera Rubin): Inference throughput improvement was measured at 4.8 times, and reinforcement learning rollout improvement was measured at 3.8 times, demonstrating cost savings and efficiency gains.
- CoreWeave Sandboxes: Sandboxes provide an isolated environment necessary for running and scaling multiple agents in parallel, enabling the full AI loop demonstration.
- Vera CPU Integration: Vera CPU is coming soon to the CoreWeave cloud, offering a tightly integrated platform where GPUs and CPUs are designed to work together.

## Practical implications

- Build engineers can now operationalize complex AI agent workflows (the AI loop) on a unified platform, reducing the need for disparate tool integrations.
- The ability to use CoreWeave Agent Lens allows teams to move beyond simple monitoring and proactively identify failure patterns in production agents.
- Model Distillation provides a critical MLOps step, enabling the creation of smaller, specialized models from large production datasets, which is essential for cost-effective deployment.
- RL Rollouts significantly accelerates the iteration cycle for reinforcement learning systems by decoupling training checkpoints from serving deployments.

## Topics

AI Infrastructure, LLMOps, Agentic Systems, MLOps, Cloud Computing, CoreWeave Forge, CoreWeave Builders Resource Center, CoreWeave Arena

Source: https://www.youtube.com/watch?v=ZdDNT6zOv8w