Agents in Production

Agents in Production A panel and demo night on what it actually takes to successfully deploy customer facing AI agents, and keep them running. Panel What we'll get into: enterprise adoption, use cases, trust and safety, and where the stack is heading over the next year. Panelists Vyas Sekar, Professor of Electrical & Computer Engineering, Carnegie Mellon University (CyLab); co-founder, Rockfish Data; Chief Scientist, Conviva. Vyas works at the intersection of security, networking, and large-scale systems, including CMU's cyber autonomy work on defending systems in the AI era. Baolu Shen, Director of Product, DoorDash, AI powered growth. Previously product leadership at TikTok and LinkedIn. Baolu is a seasoned product leader with over a decade of experience driving innovation and growth at leading Silicon Valley tech companies. Kartik Talamadupula, Distinguished Architect (AI), Oracle. Previously Head of AI at Wand AI, Director of AI Research at Symbl.ai, and Senior Research Scientist at IBM Research. Kartik's work focuses on reasoning-centric AI, large language models, and the orchestration of intelligent agents for enterprise-scale automation.Moderator - Arvind Murali Mohan, Realm Labs. Demos Satya from Scalekit Auth and actions infrastructure for AI developers. In this demo, they’ll show how developers can handle user-level authorization, token lifecycles, and tool calling without building the underlying infrastructure themselves, making it easier to build agents that can reliably take action in production. Devanshi from Headroom. Headroom Labs grew out of their open-source library, Headroom (65K+ GitHub stars, 2.6M+ downloads). They're building to make every token count for enterprise agents: compression, caching, and insight into everything an agent reads, remembers, and pays for on each model call. Shantanu from Prodigal. AI agents for loan servicing and debt collection. Details to come.Saurabh from Realm Labs. Runtime guardrails for AI agents. Their demo covers how long-horizon agent tasks fail in subtle ways, including fabricated commitments, confidently wrong answers, and gradual topic drift, then shows how reading a model's chain of thought through deep neural inspection makes those failures observable and controllable in real time.Colin from Tenki Cloud. Tenki runs the compute behind your engineering team: GitHub Actions runners, an AI code reviewer, and sandboxes for AI agents. All three drop into the GitHub setup you already have. Agenda 5:45 to 6:15 PM: Check-in and networking 6:15 to 7:45 PM: Panel and demos