The Most Interesting Thing in AI

In the last year, AI has arguably made more progress in coding than in any other domain. Its technical capabilities – harnessed through apps like Codex and Claude Code – have changed the way engineers work. Boris Cherny, Anthropic’s head of Claude Code, claims that 100% of his output is now generated by AI. But what does the rapid advance in coding tools mean for engineers and businesses?

To answer that, Nicholas Thompson, CEO of The Atlantic, sits down with engineer, author, and entrepreneur Paul Ford. The two discuss how AI has changed his work, its implications for a range of industries, and their missed opportunity to start a billion-dollar business (maybe).

(00:00) Introduction

(01:53) Writing vs. Engineering: Which profession has been transformed more by AI?

(03:23) The pitfalls of AI writing and the limits of AI agents

(04:09) How AI has changed the code development business model

(09:01) "Vibe coding": Paul Ford's personal experience building products through AI prompts

(10:47) "Death is coming" - The feeling of watching entire industries dissolve

(12:54) Where consensus programming thrives under AI assistance, and where it struggles

(15:18) Why making good software remains difficult despite faster code generation

(17:05) The restaurant review problem: The experiential engineering tasks that AI can't replicate

(21:33) Can you ship AI code without a human reviewing it?

(23:01) Will AI become self-recursive? How code runs, finds bugs, and self-corrects through iteration

(26:58) Engineers can produce 50x more code than before. What does that mean for PMs and design teams?

(29:46) Velocity vs. quality: Can AI genuinely upskill mid-level talent or just increase output speed?

(35:32) Should you teach your kids to code?

(38:36) What we learned from not building Grammarly

(42:04) Beyond agents: Incremental improvements and classic software-in-the-loop hybrid approaches
Learn more about your ad choices. Visit megaphone.fm/adchoices

In the years following the launch of ChatGPT, as concerns spread over the social and political impacts of LLMs, one person’s warnings seemed particularly dire: Yoshua Bengio’s, a scientist  and one of the “godfathers” of AI. The potential negative impacts of his life’s work weighed so heavily on Bengio that he signed his name to an open letter advocating for a pause in AI research. (The pause didn’t happen.) But recently, Bengio has found renewed optimism as he pursues a project dubbed “Scientist AI.” The pitch: What if AI didn’t care about pleasing us, and instead, like a scientist, prioritized accuracy, honesty, and probable outcomes? In a conversation with Nicholas Thompson, CEO of The Atlantic, Bengio outlines why he thinks this approach will produce better outcomes, the challenges to implementing a model that polices other (often better-funded) models, and why the age of AI– so far marked by an international arms race– will need greater international cooperation.

(00:00) Introduction

(05:00) Can we understand what's happening inside neural network vectors and attention systems?

(07:00) How ChatGPT changed Bengio’s risk assessment 

(09:37) The case for optimism (at least when it comes to technical solutions) 

(12:30) The alignment problem: AI self-preservation drives and hidden agendas emerging

(14:26) Can we train AIS to understand the world without changing it? Introducing Scientist AI

(15:52) Using Scientist AI as a guardrail to evaluate risks of actions from other AIs

(19:37) Sycophancy problem: current AIs pleasing users leads to harmful psychological effects

(22:20) The difference between Scientist AI and current value-aligned systems (Anthropic, OpenAI)

(24:06) Will AI capabilities slow down or continue accelerating beyond human intelligence?

(29:58) US-China AI race: mutual risk requiring coordination like nuclear deterrence

(31:57) UN AI advisory group with Maria Ressa: synthesizing science independently of politics

(33:18) Sovereign AI for middle powers: partnering to avoid domination by US/China

(37:54) Bengio's regret about not speaking up on AI risks earlier in his career

(40:12) How liability insurance and regulatory incentives could make safety commercially viable

(42:42) Why Europe lags in AI: capital markets and risk culture, not just regulation

(46:43) Energy consumption from AI growth and impact on fossil fuel demand

Learn more about your ad choices. Visit megaphone.fm/adchoices

Of all the potential risks and promises of AI, perhaps none are as immediately dire as this: How will it impact jobs? Will employers still need workers? What will it mean if the answer is “no?” Depending on who you’re talking to, the prospect of a future with fewer jobs is either liberating or terrifying. But for a more measured reaction, it helps to look at the data. Stanford economist Erik Bynjolfsson has done just that, drilling down into AI’s effects on employment, upskilling, output, and more. In a conversation with The Atlantic’s CEO Nichlas Thompson, Brynjolfsson goes through his studies of call centers and other AI-exposed fields, and the surprising findings that could bring some much-needed reality to our fears.

(00:00) Introduction to Erik Brynjolfsson and his work on AI economics 

(02:51) How much is free AI actually worth? 

(05:25) Why isn’t powerful AI showing up in GDP?

 (06:48) Introducing GDP-B: A new metric to capture value from free digital goods 

(07:23) Why initial AI adoption often lowers output 

(09:05) Evidence of the J-curve turning: Call centers, software, and aggregate stats 

(14:48) Will AI create more jobs or destroy them? Understanding elastic vs. inelastic demand

(19:36) Advice for students and workers: Focus on creating new value, not just efficiency 

(21:34) Why AI helps different skill levels differently in call centers vs. coding 

(25:47) The "Turing Trap": Why mimicking humans leads to substitution rather than augmentation

(30:50) Four policy recommendations: Better metrics, dynamic labor markets, and human-AI complementarity 

(37:52) "Canaries in the Coal Mine": Data showing early job displacement in AI-exposed fields 

(47:08) How higher labor costs drive automation adoption

(52:53) Fair compensation for creators: Designing incentives for the AI-content ecosystem 

(59:03) The urgent need to study the transition, not just the technology

Learn more about your ad choices. Visit megaphone.fm/adchoices

Can AI predict a person’s future? It’s a promise often made by sales teams, but the technology’s record is far from spotless. Even if it did achieve perfect foresight, a practically-clairvoyant AI might be incompatible with democracy, says Oxford philosopher Carissa Véliz. In a spirited conversation with Nicholas Thompson, CEO of The Atlantic, Véliz traces the history of predictions from ancient oracles to the modern algorithms shaping everything from criminal sentencing to insurance premiums. What are the ethics of outsourcing such consequential decisions to machines? The risks of getting it wrong are obvious. Véliz warns of the dangers of getting it right.

(00:00) Introduction to Carissa Véliz and her work on privacy and AI 

(01:29) Why "less crime" is an illusion of safety 

(02:30) How surveillance machinery enables prediction and social control 

(03:06) Defense of autonomy: why resisting surveillance protects freedom 

(04:02) Can mass surveillance ever be beneficial?

(05:36) Ring cameras and the erosion of democratic anonymity 

(07:01) Prediction versus prophecy: how forecasts shape reality 

(08:52) Job displacement predictions and self-fulfilling prophecies 

(13:15) The Gettier problem: why probabilistic AI lacks justification 

(16:06) When probabilistic AI works and when it fails 

(18:52) Do individualized health predictions defeat the purpose of insurance?

(25:24) Areas where probabilistic reasoning is inadequate: insurance, justice 

(27:48) The problems of effective altruism and utilitarian calculation 

(32:37) AI company ethics: copyright, rights, and virtue ethics 

(37:47) Can more data solve "the turkey problem?" 

(39:15) Practical privacy advice: Signal, Proton, VPN, and mindful choices

Learn more about your ad choices. Visit megaphone.fm/adchoices

Will AI destroy the world, or transform it into one of abundance? Across two books and several papers, philosopher Nick Bostrom has envisioned a range of AI futures. He joins Nicholas Thompson to discuss the ethics of how we treat AI, whether AI has sentience, and why he believes we should keep building, even at the risk of annihilation. 

Produced in collaboration with PwC.

(00:00) Introduction to Nick Bostrom and Superintelligence 

(01:56) How AI development matched Bostrom's predictions 

(04:48) Recursive self-improvement: Are we there yet? 

(07:40) Physical limits of intelligence and computational ceilings 

(09:40) Timeline predictions: Next year vs. next five years 

(11:46) Embodied intelligence: Can AI replicate motor skills? 

(14:32) Centralization vs. democratization of AI power 

(16:52) The race dynamics: One leader vs  many competitors

(19:57) AI alignment: Making systems behave as intended 

(21:37) The gap between model power and our understanding 

(23:25) "Optimal Timing for Superintelligence" paper explained 

(28:14) Swift to harbor, slow to berth: When to pause AI development 

(35:20) Moral status of digital minds and sentience 

(41:23) Building trust with potentially misaligned AI 

(44:11) Where to invest unlimited AI research funding 

(47:07) Closing: Should we say please and thank you to AI?

Learn more about your ad choices. Visit megaphone.fm/adchoices