CoreWeave has made ARIA, a coding agent that analyzes experiments, proposes next steps, and executes them with user approval, generally available inside CoreWeave Forge2.
The agent operates within Forge, the iterative AI development environment CoreWeave announced at Fully Connected 20261. Forge connects serving, observing, curating, building, and evaluating stages into a single loop designed to reduce manual handoffs between tools.
ARIA targets the experiment-iteration layer of the development cycle. Whether a team is building an agent or training a model, ARIA analyzes running experiments, proposes what to try next, and launches the proposed run once a user approves it.
ANALYSIS The human-in-the-loop approval step distinguishes ARIA from a fully autonomous system, keeping the operator in control of each iteration.
CoreWeave built Forge to address what it calls "production exhaustion," the operational complexity that accumulates after a model's first deployment. The company says AI teams today often stall between disconnected tools or lose progress in handoffs, and that Forge creates a connected improvement loop that stays open to any model, framework, or cloud a team already uses. The stated goal is to shorten the time from identifying a production issue to deploying a validated fix with every cycle.
ANALYSIS Adding a coding agent directly inside the iteration environment tightens the loop Forge was designed to accelerate, collapsing the gap between experiment analysis and the next run.