CoreWeave Launches Forge Platform to Manage Continuous AI Agent Improvement

CoreWeave Inc. has rolled out CoreWeave Forge, an integrated environment designed to bridge workflows between machine learning teams and enterprise business units deploying autonomous artificial intelligence agents.

The platform focuses on the operational challenges of maintaining autonomous models in production. Speaking to SiliconANGLE Media's livestreaming studio theCUBE during the Fully Connected event, Susanne Seitinger, vice president of product marketing at CoreWeave, outlined the system's framework for managing a continuous feedback cycle. According to reporting by SiliconANGLE (https://siliconangle.com/2026/10/08/coreweave-forge-aims-speed-ai-improvement-loop-fullyconnected/), Forge organizes agent operations into a five-stage loop: run, observe, curate, improve, and evaluate.

To support that workflow, CoreWeave introduced several platform components. Agent Lens provides visibility into live agent behavior, while Registry tracks agent configurations and model checkpoints. For optimization, the platform offers reinforcement learning rollout tooling and model distillation features. CoreWeave also announced the general availability of its AI Research and Iteration Agent, known as ARIA, which analyzes operational data to help teams identify patterns.

The software release coincides with shifting hardware demands for post-training refinement and inference workloads. CoreWeave noted that software developer Cognition AI Inc. is currently running production workloads on CoreWeave's initial deployment of Nvidia Corp. Vera Rubin NVL72 server racks.

Alongside the platform rollout, CoreWeave launched a partner network based on co-engineered integrations, aiming to deliver standardized architectures and deployment playbooks to enterprise customers.