Event Details

Coding agents are not only powerful autonomous software engineers but also serve as gateways to broader forms of digital autonomy. To bridge the gap between training and novel, real-world challenges, inference-time scaling is essential, allowing agents to effectively explore vast action spaces and complex environments.

Beginning with my work on SWE-Search, I will demonstrate how search-based techniques at inference time significantly enhance coding agent performance. Furthermore, the ability of these agents to refine their own code enables holistic adaptation in online settings, allowing agents to autonomously improve and extend their capabilities to entirely new problem domains. Together, these advancements are accelerating the transition from specialized tools to fully automated research agents.

Zoom: Join Zoom Meeting
Password: 2025

Host: DJ Ashok
POC: Maura Covaci

Speaker Bio

Antonis Antoniades, PhD ’26, is an incoming Senior Research Scientist at Google DeepMind (Gemini Agents) and is completing his PhD in Computer Science at UC Santa Barbara’s Natural Language Processing (NLP) Group. Some of his past work includes building multi-agent systems using reinforcement learning at Leela AI, Neuroformer, the first multimodal generative model for neuroscience and SWE-Search, a software engineering agent utilizing a search approach designed to emulate a human software engineer. He is also a passionate Guitar and Bouzouki player.