Eye On A.I.

What is an AI agent, really? Strip away the hype, and it's a model with access - to tools, APIs, databases, email, anything that lets it take real action instead of just generating text. That access is exactly where the risk lives, and Devvret Rishi, GM of AI at Rubrik, and former co-founder & CEO of Predibase, joins Craig Smith with a string of real-world incidents that make the case concrete: AWS reporting four major outages in 90 days after deploying coding agents, a Meta-related agent that deleted someone's emails while they were actively asking it to stop, and Rubrik's own internal pilot catching incidents that, without governance in place, would have gone unnoticed.
The conversation lays out the impossible choice most enterprises are facing right now - block AI agents and forfeit the ROI boards are demanding, or grant access and hope nothing breaks - and walks through how Rubrik's approach uses small, fine-tuned AI models to enforce plain-English security policies on every single agent action in real time. It closes on one of the most underexamined risks ahead: as agents increasingly talk to other agents to get work done, a layer of activity is forming that no human is watching, and the question of who's accountable when something goes wrong in that layer is only getting more urgent.
Subscribe to Eye on A.I. for weekly conversations with the people building and deploying the future of AI.

It costs up to $2 billion and fifteen years to develop a drug, and big pharma still fails half the time at the final stage. BullFrog AI founder, Chairman, and CEO Vin Singh joins Craig Smith with a clear diagnosis of why: the industry keeps picking the wrong drug target from the beginning, and no amount of downstream optimization fixes a fundamentally wrong starting point. Built on AI technology originally developed at Johns Hopkins' Applied Physics Lab, BullFrog has assembled a three-stage platform that cleans messy clinical data, runs causal analysis to map disease pathways, and then ranks competing drug targets using a competitive framework that removes the subjectivity most pharmaceutical decision-making still relies on.
The most striking results in this conversation come from two case studies: work with the Lieber Institute for Brain Development - analyzing thousands of post-mortem brains - that led to the identification of potential driver genes for depression, bipolar disorder, and schizophrenia in months from data that researchers had spent fifteen years studying, and a pancreatic cancer trial where BullFrog's platform identified a patient subgroup with survival rates three times higher than the study average. Vin also delivers a candid assessment of the broader AI-pharma landscape: more than 90% of AI deals in the space are missing their milestones, most companies are wrapping open-source tools rather than building genuine technology, and the shakeout between players and pretenders is already well underway.
Subscribe to Eye on A.I. for weekly conversations with the people building and deploying the future of AI.

After two years of AI pilots, enterprises are finally diagnosing what went wrong, and the answer keeps coming back to data. Alberto Pan, CTO of Denodo, joins Craig Smith to walk through the findings of the company's AI Trust Gap Report: a survey of 850 enterprise data leaders that reveals the dominant failure modes of enterprise AI agents are almost never the model's fault. They're caused by stale data, missing context, and inconsistent semantics across the hundreds of data sources agents need to access to do real work.
Pan explains why traditional data warehouse and lake house architectures - built for analytics, not real-time decision-making - are creating an invisible ceiling on AI performance, and how Denodo's logical data management approach lets agents query data where it lives without centralizing it first, while enforcing consistent governance across every source in one place. The conversation also identifies two specific traps most organizations fall into as they try to scale AI - over-centralizing data into a single system, or building custom ad hoc data layers for every agent - and why both approaches collapse in a multi-agent world where agents need to cooperate, share context, and work from a common semantic foundation.
Subscribe to Eye on A.I. for weekly conversations with the people building and deploying the future of AI.

What if consciousness isn't a byproduct of complex brains, but a fundamental feature of reality itself, present, in some rudimentary form, all the way down to electrons and quarks? Philip Goff, a philosopher at Durham University and one of panpsychism's leading contemporary advocates, joins Craig Smith to make that case, arguing that modern science's founding move - separating the mathematical world physics studies from the subjective experience we know only from the inside - solved one problem while quietly creating another we've never resolved.
The conversation inevitably turns to AI: could a large language model ever be conscious? Goff's answer is a careful, well-reasoned no, not because he thinks consciousness is magical, but because his framework treats it as something closer to the physical substance of reality than an abstract computation, making him skeptical that anything resembling current AI architecture could cross that threshold. Along the way, he tackles one of the genuine open mysteries in his field: if natural selection only cares about behavior, why did evolution bother making us conscious at all, and what would it even mean to find experimental evidence for an answer.
Subscribe to Eye on A.I. for weekly conversations with the people building and deploying the future of AI.v

Eighty percent of lung cancer cases are diagnosed too late, not because the signals aren't there, but because nobody was looking at the right moment. Prashant Warier, co-founder and CEO of Qure.ai, joins Craig Smith to explain how his company is changing that using a tool most people already encounter: the routine chest X-ray. Cure's Lung Nodule Malignancy Risk Score - validated in the CREATE study - analyzes X-rays people get for unrelated reasons, identifies high-risk nodules, and flags which patients need follow-up CT scans. The result is a detection rate of 54 positive patients out of 100 flagged as high-risk, compared to the 2 out of 100 found by standard CT screening programs. That's not a marginal improvement. That's a different category of outcome.
The conversation covers the full landscape of where AI diagnostics actually stands today: the 15 million TB screening X-rays that Cure reads autonomously every year across 70 countries with no radiologist in the loop, because in many of those countries there are only two radiologists for the entire nation; the 26 FDA clearances and 200-plus published studies that underpin the company's clinical credibility; and the regulatory barriers that currently prevent patients from uploading their own scans and getting an AI read directly. Warier also makes his sharpest prediction: within 5 to 10 years, primary care will be AI-first, the first conversation you have when something feels wrong won't be with a doctor, it will be with an AI. Based on what Cure is already doing at scale today, that timeline is harder to dismiss than it might sound.
Subscribe to Eye on A.I. for weekly conversations with the people building and deploying the future of AI.