Ep. 3 | What Orange’s Chief AI Officer Learned from General Magic | Steve Jarrett

John Giannandrea, who ran AI at Google and Apple, and Steve Jarrett, former Chief AI Officer at Orange, talk with us about AI hype, regulation and why the best technology comes from values, not money. It’s a conversation about hype, accountability and humility from two General Magic alumni and AI pioneers. In this episode * 00:29 – Introducing the guests: John Giannandrea and Steve Jarrett, two General Magic alumni at the center of the advent of AI * 01:31 – Steve’s opening question: what has surprised John most about AI, compared with the smartphone and the web? * 02:03 – John on AI’s momentum, cyber risk, and Amara’s law: overhyped in the short term, underappreciated in the long term, with a story from the Netscape era * 04:30 – Apple’s stance on privacy: why personal data shouldn’t be used to train models, and why the most useful AI works over your most personal data * 06:11 – What runs on your device and what runs in the cloud, and the “quiet” intelligence behind iPhone focus modes * 07:38 – Mike on Marc Porat’s prediction of a superintelligence in your pocket. John on the AI tutor, Danny Hillis’s essay “Aristotle,” and why AI remains “jagged” * 09:09 – Hersh on hallucinations in the legal field. Power tools don’t make carpenters go away, and LLMs can’t do your taxes * 10:24 – Guardrails and regulation: Steve on the EU AI Act, what worked and what stifled open-source innovation * 13:33 – John: transparency matters more than regulation. Model cards, independent benchmarks, and why the labs are “marking their own homework” * 15:41 – Hersh asks what’s inherently dangerous about AI. John: we don’t understand why these models work * 16:55 – Mike’s seat-belt analogy, what would make industry act, and John’s idea of a “nutrition label” for models (Common Sense Media’s benchmark work) * 18:24 – Steve on how Orange managed risk, and the split between Western and Chinese AI ecosystems * 19:40 – John disagrees: China has the most regulated AI ecosystem in the world. Steve on open-weight models, bad actors and an AI arms race * 21:27 – Hersh asks about the ethos of innovation: how has the philosophy behind your work evolved? * 22:17 – John on values in Silicon Valley: What the Dormouse Said, the origins of the internet, Vint Cerf, and why General Magic was in that tradition * 24:04 – Mike asks whether the valley has grown “rancid” with the tech broliarchy. John: some parts, not all. Musk’s interplanetary vision, the SETI Institute and Fermi’s paradox * 25:32 – Steve on natural language as the new programming, and Bill Atkinson’s HyperCard * 26:21 – John on technology adoption: the fax-machine cycle, AI as a “spreadsheet moment” * 28:16 – Can today’s architectures reach human-level intelligence? World models, Yann LeCun, causality, and why a child can learn from one example * 30:56 – John: “the models are the most expensive they’ll ever be,” and why he expects a 100x to 1,000x efficiency gain * 32:01 – Hersh asks what tendencies AI promotes. Steve on the generational question of how young people will learn the hard way * 33:26 – John: the case for being a technological optimist, and mobile access versus sanitation * 35:29 – Has AI changed how John works? Why there’s no evidence yet that AI creates novel thoughts, and AlphaGo * 37:19 – Steve on AI teaching, a kid in the Congo, and John’s Feynman story * 38:33 – Mike on the allocation of capital. John: “I’ve seen this movie before.” The dot-com fiber build-out, data centers, and Chewy, Instacart and Webvan * 40:16 – Back to General Magic: John’s seminal moment seeing NCSA Mosaic, “worse is better,” and why he quit * 42:55 – The through line of John’s career: working on things people are skeptical of * 44:16 – Has Silicon Valley lost that ethos? John’s answer: tell the full history, and the talk he would like to give * 45:26 – Steve on why the next generation is inspiring * 46:03 – “Hubris and humility. You need both,” and General Magic car stories * 47:07 – Hersh’s closing thanks About J

Our guest, Eric Goldman, professor and Associate Dean for Research at Santa Clara University School of Law, was dubbed The Internet’s Last True Believer in a Santa Clara Magazine profile in 2023. Eric Goldman has spent three decades in internet law as a Cooley associate, a general counsel, and now a professor and Associate Dean at Santa Clara University School of Law. He is one of the leading authorities on Section 230, the 1996 law that says websites aren’t liable for their users’ content. He is also one of the last people still making the case for the open internet he fell in love with in the early 1990s. Mike and Hersh came to this conversation with strong views on age gating, platform liability and regulating social media. Eric took them apart, politely and with rigor. He explains what Section 230 really does, why he doesn’t buy the claim that platforms are “intentionally designed to cause harm,” and the five ways age authentication online is different from checking ID at a liquor store. He argues that regulators reach for the light switch (ban it or leave it) when they should be looking at the dimmer, and he warns that the real threat to generative AI is the government deciding what chatbots are allowed to say. It is also a conversation about how the General Magic generation’s faith in technology has aged. Eric talks about how his own view has changed: a clearer sense of why the rule of law matters, a lost presumption of good faith in tech companies, and a classroom in which half his Gen Z students would turn shut internet down if they could. Along the way: what he learned from the Trust & Safety community he helped build, why he bristles at the term “surveillance capitalism,” and the bumper sticker on his desk. In this episode * 01:10 – What is Section 230? Why Congress passed it in 1996, and how it let services host conversations without being liable for them * 03:05 – The printing-press analogy, and why publishing other people’s content was risky for centuries * 04:10 – The two choices without Section 230: a cesspool or publishers exit * 05:01 – Mike’s challenge: after 30 years, is it time to retire Section 230? Platforms, harm by design and the Meta trials * 06:23 – Eric pushes back: Section 230 isn’t what is holding back those lawsuits, and “product marketing” isn’t the same as designing to harm * 07:30 – The tobacco comparison, teenagers and body image, and Meta facing competing groups of users * 09:03 – Can a technology be neutral? Hersh’s “kinder, gentler gun,” and how to think about AI’s likely social effects * 10:18 – Eric’s scissors example, the limits of what technology makers can fix, and why the answer is “why not both?” * 12:54 – “Generative AI Is Doomed”: why the biggest threat is partisan control of what chatbots say * 15:53 – Mike asks about age gating: we restrict cars, voting and alcohol by age, so why not AI? * 16:21 – Eric’s five differences between online and offline age authentication: privacy, security, friction, cost and surveillance, plus the First Amendment * 21:07 – What’s on the other side of the scale? Teaching kids to cross the “digital street” (danah boyd’s analogy), and a detour into jaywalking laws * 24:01 – Can anyone test Section 230’s absence? What the rest of the world’s internet tells us * 26:14 – “Content Moderation Remedies”: think of a dimmer, not an on/off switch, and why governments reach for bans * 29:10 – Financial consequences inside platforms, and how demonetization works * 31:10 – Has Eric changed his mind? The rule of law, the end of presumed good faith in tech companies, and General Magic * 33:49 – His Gen Z students: half would switch off the internet, and many arrive anti-AI * 35:36 – How Eric actually uses generative AI, and Mike’s PowerPoint experiment * 37:22 – Hersh’s profanity rule and the bumper sticker on Eric’s desk * 38:31 – Why Eric doesn’t do much policy work anymore, and the Trust & Safety Professional Association * 39:44 – His nightmare outcome: government

At General Magic, we naively believed that if we built elegant tools and put them in people’s hands, the rest would take care of itself. Tony Fadell was there at the beginning, a young engineer with an uncommon gift for making technology disappear into the life of the person holding it, and he went on to help build the iPod, the iPhone, and Nest. So when he says that every barrier to our vision has since been knocked down—“the pipes are open and they’re flooding right to your eyeballs and your fingertips”—he is describing a demolition he took part in. He does not flinch from that. He calls what followed a frictionless flow of everything, for better and for worse, and he declines to pretend the worse was an accident. What distinguishes this conversation is where Tony locates the failure: not in a few bad actors, but in structure. Cars became safe when government forced the issue and people had stopped dying; a hands-off posture toward AI and social media is an abdication, and the missing ingredient is political will, in a system where money decides whose voice is heard. He sees the next trap plainly. Consumers will not pay $30 a month for an AI assistant, which means the business model will drift, as it did before, toward ads and data, and we will once again be the product. He also says what few of his peers will say aloud: that the centi-billionaires who now set the tone for an entire industry have forgotten the society and its ethical norms which made their fortunes possible, and that Fortune 500 CEOs, once answerable to their employees and to the outside world, now seem answerable to no one. The phrase that stuck with us is cognitive surrender. Tony describes people who supervise a hundred software agents but end up like quality-control workers on an Amazon conveyor, moving bins from one machine to the next, at the speed of the system rather than the speed of humans, until the craft is gone and burnout is all that remains. There is grief in that image—the ache of watching work that once carried a person’s whole attention reduced to throughput. His solution is modest and historically grounded: Microsoft once sent salespeople to sit across from customers who disliked the company, because it is very hard to hate someone face to face. Build technology that gathers people in person and amplifies what they say. Put the phone down at dinner. Back only companies that make the world a better place. Tony admits he feels alone in this, a humanistic technologist in a Valley that has moved on. The unedited episode of our interview with Tony is on YouTube.

Before she was Employee #12 at General Magic, Megan Smith grew up in Buffalo, studied mechanical engineering at MIT, and was working in Tokyo on multimedia projects for Apple Japan when a chance meeting with Bill Atkinson brought her into the company building the first mobile communication devices. She went on to an eleven-year run at Google — leading the acquisitions of Google Earth and Google Maps — before President Obama appointed her the third Chief Technology Officer of the United States in 2014. In this conversation with Mike and Hersh, Megan makes the case for an optimism she’s carried since General Magic: that the world’s hardest problems — climate, AI safety, equity in tech — don’t get solved by any one person or technology, but by connecting the right people to each other. She traces that thread from General Magic’s apprentice-journey-mastery culture through her civic-tech company Shift7 and its “Scout & Scale” approach, the UN Sustainable Development Goals Solutions Summit’, and a career-long rule borrowed from Diane von Furstenberg. Along the way, she reflects on the erasure of women like Ada Lovelace and Grace Hopper from computing history, revisits the safety-first design philosophy behind Telescript’s agentic design in the 1990s, and debates what it would take to build “kindness by design” into today’s AI. In this episode * 01:29 – “Go where there’s tons of us, and go where you’re rare” — the 2015 MIT commencement advice behind her civic-tech focus; Shift7 and her board work (Think of Us, Planet Read, Earth Conservation Corps, MIT, LA28) * 04:31 – The “three-layer cake” of policy, law, and tech innovation; President Obama, the Army Corps of Engineers, and the history of science advisors in government * 06:57 – Why “harder than self-driving cars” problems don’t get tech’s top talent * 09:02 – What General Magic — Steve Jobs’s original Mac team gone mobile — taught her about apprentice-journey-mastery; the original vision for a “dashboard” device vs. today’s attention economy; the Google Earth/Maps acquisitions and the Mapmaker crowdsourcing story * 14:53 – Weighing models for AI regulation * 15:17 – Dr. Joy Buolamwini, Gender Shades, and the Algorithmic Justice League; ROOST and open child-safety tools * 17:12 – Telescript’s “should we?” vs. AI’s “can we?” — picking up the design-philosophy thread from the Rich Miller episode * 18:20 – AI companions, “aggregated” intelligence, and Megan’s “Scout & Scale” approach * 23:00 – The UN Sustainable Development Goals Solutions Summit: 800 submissions from 131 countries * 25:45 – Remembering mentor Wendell Sanders and General Magic’s first hardware team * 27:44 – Stanford, venture capital, and the fight for the university’s civic mission * 30:11 – The American Research Enterprise, teaching kids to “cook” with electronics, and pioneering environmental scientist Ellen Swallow Richards * 35:24 – Early AOL, Steve Case, and the quilters who met through chat rooms; her own time running PlanetOut * 41:17 – “Kindness by design,” Pope Leo XIV, and Cathy O’Neil’s Weapons of Math Destruction * 41:58 – The Blueprint for an AI Bill of Rights and colleague Alexander Macgillivray * 43:22 – The grassroots data-center backlash, and watching Texas’s ERCOT grid run on solar * 46:36 – Women erased from tech history: Bletchley Park, ENIAC’s six programmers, Grace Hopper, and Ada Lovelace’s hidden signature “A.A.L.” * 50:08 – Her trajectory from Employee #12 at 23 to U.S. CTO * 50:20 – Diane von Furstenberg’s rule: Connect two people every day

What do you get when you cross a New Mexico cowboy with the early days of email and telecommunications? Answer: Rich Miller, one of the OGs of email and, with his colleague and close friend Jim White, the force behind General Magic’s communications language, Telescript. Amidst today’s talk of agentic AI, Rich Miller recollects the original agentic concept and how, from the start, safety considerations like identification, authentication, and authorization were front and center rather than relegated to the “we’ll fix it in post” attitude rife among today’s purveyors of a new class of agents. Rich is measured and thoughtful yet does not mince words. “The notion of complete autonomy of agents is a recipe for a disaster and how big a disaster is the only question.” He offers a nuanced approach for heading off the worst possibilities, discussing considerations across economics, public policy, legal frameworks, and more. So, saddle up for a deep dive into the deep issues at the heart of whether a technology will ultimately serve us well or run amok. Show notes Before Rich Miller was CEO of Telematica and Executive Chairman of Provenant Data, he spent roughly fifteen years building electronic mail into a real industry — running early ARPANET messaging work at Infomedia Corporation, serving as Chairman of the Electronic Mail Association, and representing the U.S. as a delegate to the international CCITT standards body. That’s the résumé he brought to General Magic in 1990 as VP of Strategic Partners, where he and his longtime collaborator Jim White architected Telescript, the agent-based network technology GM built years before “agentic AI” was a phrase anyone used. Telescript didn’t just let software agents travel across a network to act on your behalf — it fenced them in: agents needed explicit credentials, defined authorities, and permits before they could do anything, with the guardrails built in from day one rather than bolted on after the fact. In this conversation, recorded after the passing of Jim White — Telescript’s co-architect and, by Rich’s account, its moral center — Rich joins Michael Stern and Tom “Hersh” Hershenson to walk through how Telescript worked, why AT&T forced General Magic to build for safety before scale, and why he believes the AI industry is now rebuilding what Telescript already had, minus the accountability — plus what he’s building now at Provenant Data to give AI a memory of where its facts actually came from. In this episode * 01:35 – Rich Miller’s path: ARPANET messaging, chairing the Electronic Mail Association, and building email into an industry — plus the MagicCap vs. Telescript culture divide inside General Magic * 05:33 – Jim White — Telescript’s co-architect and moral center, remembered in detail * 08:47 – What is Telescript? The PostScript analogy for sending intent to an “engine” * 10:49 – Software agents in the wild — the 1994 travel-booking use case * 13:35 – Deterministic safety-by-design vs. today’s constrain-after-the-fact approach * 18:11 – The “architect of spam” & lessons from early email standards * 19:35 – Constitutional AI, alignment, and human-in-the-loop for autonomous systems * 23:24 – AT&T’s enforced guardrails as a model for accountable infrastructure * 30:59 – Governance, treaties, and who pays for AI’s externalities * 36:01 – Provenant Data and the case for data lineage and provenance in AI * 44:33 – Closing reflections & remembering Jim White About Rich Miller Rich Miller spent roughly 15 years building electronic mail into an industry — at Infomedia Corporation on early ARPANET messaging, as Chairman of the Electronic Mail Association, and as a U.S. delegate to the CCITT standards body — before co-founding Rapport Communication with Jim White and Ted Myer, where Telescript was conceived. He joined General Magic in 1990, building the AT&T PersonaLink service and the NTT Japan partnership. Today he is CEO of Telematica and Executive Chairman of Provenant Data, writing