Podcasts

Anthropic flew a Vedanta monk to San Francisco to help think through how Claude should tell right from wrong. That monk is Swami Sarvapriyananda, head of the Vedanta Society of New York. What fascinates me is how naturally this moment fits into a trajectory I’ve been watching for decades. I’ve been interested in the 𝗶𝗻𝘁𝗲𝗿𝘀𝗲𝗰𝘁𝗶𝗼𝗻 𝗼𝗳 𝘁𝗲𝗰𝗵𝗻𝗼𝗹𝗼𝗴𝘆, 𝗰𝗼𝗻𝘀𝗰𝗶𝗼𝘂𝘀𝗻𝗲𝘀𝘀 𝗮𝗻𝗱 𝗩𝗲𝗱𝗮𝗻𝘁𝗮 for many years. In 2024, I explored that intersection in my b

Good brakes are what let you drive fast with AI. That is one of the clearest messages in Boston Consulting Group (BCG)’s latest research. Almost half of companies are now generating value from AI. The real story is the growing gap between those that are merely adopting it and those turning it into business impact. Strategic clarity + applied AI = transformative impact. Companies that excel at both generate 5X more AI value than those that do not. A few highlights that caught

(𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲 + 𝗔𝗜) × 𝗠𝗲𝗻𝘁𝗮𝗹 𝗠𝘂𝘀𝗰𝗹𝗲 = 𝗟𝗲𝘃𝗲𝗿𝗮𝗴𝗲 I use AI constantly, including frontier models and tools that are still pre-release. It is an extraordinary privilege to experience this technology while it is evolving so rapidly. For me, 𝗔𝗜 𝗵𝗮𝘀 𝗯𝗲𝗰𝗼𝗺𝗲 𝗮 𝗳𝗼𝗿𝗰𝗲 𝗺𝘂𝗹𝘁𝗶𝗽𝗹𝗶𝗲𝗿. Years of working across technology, AI, data, finance, business, research and writing have given me a large collection of mental models, questions, intui

Why are the companies warning us about AI also racing to build more of it? And what is all this intelligence doing to us? This week: → OpenAI paused its latest model over safety concerns, then launched always-on agents ("Dots"), ChatGPT-6.1 Sol with near-frontier intelligence at a fraction of the cost. → Anthropic, the lab most associated with AI safety, filed an IPO prospectus: revenue up 12X to $4.6B, +$500B in planned infra commitments, targeted valuation +$2T. → The Wall

Opus 5.5 is incredible. Kudos to Anthropic. I would call this "competent intelligence." Let me explain. For the last few years we've had useful intelligence: a draft to rework, a snippet to debug, an answer to double-check. Competent intelligence == hand it direction and get back finished work, across multiple tools, ready to ship in real-world circumstances. The data backs this up. OpenAI built GDPval to test AI on real work including tasks from 44 occupations across the 9 i

AI agents are now calling customer service. And increasingly, AI may be on BOTH sides of the call. Business Insider reports that AI agents are already sitting on hold, negotiating bills, canceling subscriptions, and calling insurance companies on behalf of customers. And, Forbes reports people are already using agents to save real $$. That points to a much bigger shift. I have been writing about this for the last five years: Substitution creates capacity. Reinvention determin

AI is forcing us to rethink what education is actually for. We are seeing very different perspectives emerge on the same question -> 1\ Restrict it. New York City is banning student-facing generative AI through 8th grade for the 2026-27 school year. Los Angeles has imposed a moratorium on student use of generative AI on school-issued devices. Many schools have banned AI-use. The logic is that students need to wrestle with hard problems themselves to build judgment, critical t

Meta unveiled a new stack of AI products: 1) Muse, a fast-growing personal AI agent inspired in part by the OpenClaw model, 2) lighter VR glasses that create a private spatial workspace, 3) audio-only AI glasses; and Muse Charm, 4) a tiny keychain device that puts the agent one tap away. The individual products are interesting BUT the direction of travel far more important. OpenAI, Anthropic and Google are all moving toward increasingly capable agents that can use software, a

Anthropic just launched Claude Opus 5.5, another reminder that the AI model race is still moving fast. There’s a growing sense of “model fatigue” with AI. Every few weeks (sometimes days) brings another model, another benchmark, another claim of being the smartest. What matters now is whether these systems actually become more useful in the real world. Opus 5.5 is a meaningful step in that direction: it outperforms Opus 5 and Fable 5.1 on nearly every benchmark, leads in agen