Atul Thapliyal

Talk Title:

From Models to Agentic Systems: How They Act, Fail, and Earn Trust

Abstract:

The topic explores the transition from traditional AI models to agentic systems that can reason, act, coordinate, and interact with external environments, emphasizing that increased autonomy also introduces greater responsibility and new failure surfaces. It examines how failures can emerge across context, retrieval, orchestration, tools, controls, evaluation, and observability rather than solely from model reasoning. The presentation introduces Context, Control, Coverage, and Evidence (C³E) as a practical design lens for diagnosing failures and strengthening the reliability of agentic systems. Ultimately, it emphasizes building trustworthy AI agents whose actions are observable, testable, governed, and supported by auditable evidence, while preserving appropriate human oversight.

Profile:

Atul Thapliyal is a Senior Managing Consultant in GenAI Strategy & Implementation at IBM Consulting, based in Denver, where he leads enterprise AI transformation and AIOps programmes for Fortune 500 clients. He is an Enterprise Design Thinking Coach and a named inventor on multiple patents. Alongside his client work, he designs and instruments a personal multi-agent environment — work that informs his current focus on why agentic systems fail and what makes their actions observable, testable and governable.
He is an active mentor, hackathon judge and university speaker.