Sign In
Register

Partnership opportunities

Secure your seat

Call to action
Your text goes here. Insert your content, thoughts, or information in this space.
Button

Back to speakers

Sayantan
Ghosh
Senior Engineering Manager
LinkedIn
Sayantan Ghosh is an award-winning senior engineering leader with deep expertise in building large-scale AI and Data platforms that power products used by billions globally. With leadership roles across Meta, Uber, LinkedIn, and eBay, he has driven some of the industry’s most influential machine learning and data infrastructure initiatives, including Uber’s Michelangelo ML Platform, Meta’s FBLearner ML Platform and LinkedIn’s Feed Data Platform, which power multi-billion dollar lines of business like Uber Eats, Instagram Reels, FB Newsfeed, Facebook Marketplace etc. Sayantan holds a widely cited US patent, is a published author, serves on program and review committees of leading international conferences and is a frequent invited speaker at international venues. An alumnus of IIT Kharagpur, a Senior IEEE Member, and an IETE Fellow, Sayantan has mentored several engineers and managers through pivotal career transitions.
Button
01 December 2026 12:00 - 12:30
Agentic AI doesn't have an intelligence problem, It has a governance problem
Giving an agent more intelligence is the easy part. Giving it clear limits on what it is allowed to do is something entirely different. As agents move from answering questions to taking actions across real systems, most of the failures teams run into come from missing guardrails rather than weak models, whether that is unclear permissions, no audit trail, or nobody accountable when an agent gets something wrong. So if the model isn't the bottleneck, what does it actually take to make agentic AI reliable enough to trust in production? This session makes the case that the next leap in agentic AI will come from governance, not intelligence, drawing on hands-on experience building AI and data platforms at hyperscale. Through a practical look at how governance can be designed into agent systems from the start, rather than bolted on after something breaks, it covers the controls, checkpoints and ownership models that let teams scale autonomy without losing accountability. Key takeaways: → Why stronger models don't fix agents that act without clear boundaries → How to build permissions, audit trails and human checkpoints into agent workflows from day one → What accountability looks like when an agent makes a decision nobody explicitly approved