LLMs are the most overrated part of AI - Sanjay Rakshit

About This Episode

Bhairav sits down with Sanjay Rakshit, now at Poppulo, for his third or fourth appearance on the show.

The conversation was sparked by a quote Sanjay gave at DSE (the digital signage industry event run by Invidis): "LLMs are the most overrated part of AI." They unpack what that actually means, and dig into agents, agentic systems, context engineering, data fidelity, AGI, robotics, and the future of engineering jobs.

What We Cover

- What Sanjay meant by his "LLMs are overrated" quote, and the DSE event where he said it - The difference between an agent and a chatbot: agency, planning, and achieving an outcome versus reactive output - Where LLMs actually fit inside an agentic system (as one tool among many) - Why context engineering, not the LLM call, is where most of the real work happens - Data fidelity versus data quality, and why it matters more than raw data volume - The mathematics underpinning AI, and why understanding first principles beats following influencers - Quantum computing's potential impact on AI going forward - AI, jobs, and the "power tools" analogy for engineering productivity - Sanjay's take on AGI, and why he remains sceptical - Robotics versus AGI, and his work with the Bayes Institute in Scotland

Key Takeaways

- An agent is any entity that can think, plan, and act to achieve an outcome; a chatbot just responds - Roughly 70% of agentic system work is context generation and systems engineering, not the LLM call itself - Data fidelity (relevance to the use case) matters more than raw data quality - AGI scepticism rests on the argument that the brain's capacity may be too vast to model - AI is compressing the mechanical effort of coding, freeing engineers to focus on solving the actual problem

Links

Connect with Sanjay Rakshit on LinkedIn: https://www.linkedin.com/in/sanjay-rakshit/