Tim Sweeney

Tim Sweeney was identified in the 2026 AI Engineer World’s Fair biography as a Principal Engineer at Weights & Biases by CoreWeave. His main-stage session, “Closing the Loop: An Autonomous AI Research Agent,” explored autonomous AI research agents and the feedback loops behind them. In his introduction, Sweeney reported having a master’s in machine learning and reinforcement learning from Georgia Tech and previously serving as PM of Twitter’s ML stack.

Sweeney demonstrated ARIA iterating on code and hyperparameters, launching GPU experiments through W&B Launch, summarizing strong training runs, identifying experimental patterns, and creating reports and workspaces with built-in charts. He explained how the team improves the agent by logging traces to Weave, reviewing conversations with humans, and using live LLM judges to flag user frustration and low-quality responses. Those observations become YAML evaluation tasks that check correctness, insight quality, and tool-call efficiency; roughly 200 tasks run nightly to inform release decisions. His approach emphasizes observing agent behavior, treating evaluations as continuous integration, retaining human judgment, and improving domain context and tools before adding complexity to the agent harness.