CoreWeave Forge targets faster agent AI loop

CoreWeave's Forge arrives as enterprises move agents from pilot projects into production, where the work of running, observing, evaluating and improving them does not stop. Seitinger described the platform as a way to connect disparate pieces so different teams can talk to each other more effectively. The offering is built around a repeatable cycle: run, observe, curate, improve, evaluate and repeat. Several of the new capabilities target specific pain points in that cycle. Agent Lens lets teams inspect what agents are doing, Registry checks model checkpoints and agent configurations, while RL Rollouts and model distillation handle improvement. ARIA, the company's AI Research and Iteration Agent, is now generally available and is meant to help users spot patterns. Seitinger said speed of learning is the measure that matters most, tying it to faster shipping and faster value. She also pointed to changing workload mix as a reason the timing matters, noting that Cognition AI Inc. is already running production workloads on CoreWeave's first Nvidia Corp. Vera Rubin NVL72 racks. Alongside the platform, CoreWeave introduced a partner network based on tested, co-engineered integrations, which Seitinger described as a way to give customers more recipes, playbooks and use cases so they do not have to repeat the same learnings.