ECE Distinguished Lecture Series- Claudionor Coelho, Senior Fellow for AI at Majestic Labs

Beyond Foundation Models: Are LLMs in the Road to AGI, or Are They Your Next Pet Rock?

Bio: Claudionor N. Coelho Jr. is a Senior Fellow for AI at Majestic Labs, focusing on AI-inspired simulation of future architectures and workloads, and an Invited Professor teaching deep learning and natural language processing at Santa Clara University. He has been actively building AI Copilots and Multi-Agent systems since 2016, having developed several major systems to date.

Previously, Claudionor served as Chief AI Officer at Zscaler and Advantest, and as VP/Fellow of AI at Palo Alto Networks, where he led initiatives in neuro-symbolic AI, AIOps, and True Zero Trust cloud solutions. He was also a consulting member for the World Economic Forum, co-authoring recent reports on AI Agents and cybersecurity.

During his time in machine learning research at Google, he developed QKeras, an automatic quantization and model compression technology utilized by CERN and featured in Nature Machine Intelligence in 2021. His earlier career includes executive roles at Cadence, Jasper Design Automation, and Synopsys, where he led an international engineering team of 250 personnel and managed business development in South Korea.

A pioneer in the Latin American AI ecosystem, he founded Kunumi, Brazil's first deep learning company, in 2014, and later invested in Hyperplane. Both startups were recently acquired by Bradesco and Nubank. He was also responsible for expanding Cadence and Jasper Design Automation into South America and opening the first ACM Programming Contest site in the Southern Hemisphere.

Claudionor holds a Ph.D. in Electrical Engineering and Computer Science from Stanford University, an MBA from Ibmec, and an M.Sc. and B.S. (summa cum laude) from the Universidade Federal de Minas Gerais.

Abstract:

Never in the history of humankind has a new technology moved as fast as AI, Large Language Models, and Agents. While the tech industry is currently captivated by this rapid advancement, beneath the hype, we must ask: are these systems truly stepping stones to Artificial General Intelligence, or simply modern-day "Pet Rocks", that are obedient algorithms that sit, stay, and rely heavily on human confirmation bias to appear intelligent? This talk dismantles the illusion of conscious AI and explores the hard realities of transitioning from basic foundation models to complex, multi-agent systems.

To navigate this transition, we need to take a step back and think beyond the marketing hype. It has never been more important to return to the engineering and computer science basics that people are currently forgetting. We are building complex engineering systems where AI is an important part of the overall architecture, rather than a standalone magic bullet. While Agentic AI promises radical simplification and autonomous workflows, it introduces profound challenges in both reliability and security. We will explore the mathematical limits of multi-agent systems, demonstrating how and when these systems will fail. Furthermore, we will examine the emerging security nightmares introduced by "Bring Your Own Agent" (BYOA) architectures and protocols like MCP, which can expose highly confidential corporate data to unauthorized internal and external exploitation.