
Francesco Bonacci is co-founder and CEO of Cua (YC X25) and former Engineer at Microsoft. Cua builds the sandboxes, environments, and reinforcement-learning data that frontier labs use to train and evaluate agents that operate desktop and mobile applications. Its open-source framework lets developers spin up computer-use agents in a few lines of code, and its benchmark, Cua-Bench, alongside a catalog of thousands of cross-platform RL environments, is used by leading AI teams to measure real GUI task performance. Francesco works closely with research teams across the agent ecosystem on environment design, grounding data, and evaluation.
Using publicly available information we constructed an analysis to help you get a feel for this speaker before deciding to attend their session.
Bonacci is the founder/CEO of Cua, the open-source infrastructure for computer-use agents (18.7k+ GitHub stars), and co-authored Microsoft's Windows Agent Arena benchmark (ICML 2025); expect a hands-on, internals-deep session on how to actually run and evaluate agents that drive full desktops (macOS/Windows/Linux), sandboxing trade-offs, and benchmarking.