📬 This is the companion episode guide to AI Was Supposed to Accelerate Diligence. It's Killing Deals.
Subscribe to get the full briefing →
Episode Guide: AI Was Supposed to Accelerate Diligence. It's Killing Deals.
Companion to the Sunday, April 19, 2026 edition of Transformation Brief: AI & Technology
This edition covers 12 episodes spanning AI Populism, Enterprise AI Adoption, Public Perception of AI, Trust in AI Leadership, Productivity Gains. Below you'll find detailed breakdowns of every episode referenced in today's briefing — including key guests, standout quotes, and links to listen.
Beyond The Prompt - How to use AI in your company — "Nobody Is Getting New Manager Training for Their AI Team - with Dan Klein, UC Berkeley"
Runtime: 63 min | Host: Jeremy Utley & Henrik Werdelin | Guest: Dan Klein (UC Berkeley)
For: CEOs and CTOs grappling with AI reliability and team training. This episode dives into why AI "hallucinations" aren't always a bug, the "jagged frontier" of AI capabilities, and the critical need for new managerial skills to effectively work with these systems.
Dan Klein, a UC Berkeley professor and CTO at Scaled Cognition, unpacks the fundamental nature of large language models: they are built for fluency and confidence, not inherent truth. This leads to what he calls the "jagged frontier" – areas where AI excels versus where it falls flat. Understanding this distinction is crucial for effective human-AI collaboration, especially as companies realize they need new "manager training" for their AI teams.
"The systems we've built, really, they are fundamentally systems designed to produce outputs indistinguishable from the truth. That's different than outputting correct answers. They're fluent, they're confident. The parts we do understand look correct. We assume that everything else is correct. And that's not always true." — Dan Klein
Connects to: Lack of AI literacy investment by enterprises, Hallucination as a feature vs. bug in AI, AI fluency vs. truth/accuracy
Practical AI — "Open Source Self-Driving with Comma AI"
Runtime: 46 min | Host: Daniel Whitenack & Chris Benson | Guest: Harald Schäfer (Comma AI)
For: Technical leaders and investors in autonomous systems. Explore the pragmatic, open-source approach Comma AI is taking to self-driving, focusing on end-to-end learning and innovative simulation techniques, and the surprising challenges in low-level controls.
Harald Schäfer, CTO at Comma AI, explains their OpenPilot, an open-source autonomy stack, and how they train models from raw sensory input to vehicle actions. The conversation reveals their innovative use of machine learning simulations (akin to Sora) and the paradox that while ML drives much of their progress, classical optimization still reigns supreme in certain low-level, real-time control systems. It's a candid look at building useful, shippable features while pushing the frontier of autonomous driving.
"Our mission is to solve self driving cars while shipping intermediaries... we want to make progress and at the meantime be able to ship useful features." — Harald Schäfer
Connects to: Python for machine learning development, low level controls in robotics, end-to-end training (self-driving)
The AI Daily Brief: Artificial Intelligence News and Analysis — "AI Populism Turns Violent"
Runtime: 32 min | Host: Nathaniel Whittemore | Guest: Nathaniel Whittemore
For: Executives, board members, and policymakers tracking AI's societal impact. This episode unpacks the unsettling rise of AI populism and violence, linking it to economic anxieties and the AI industry's own rhetoric, and urging for a restored democratic channel for governance.
Nathaniel Whittemore connects recent violent acts targeting AI figures and infrastructure to a deepening "AI populism," fueled by perceived economic inequality and a sense of disenfranchisement. He argues that the industry's own "X-risk" rhetoric, combined with lobbying against regulation, paradoxically contributes to public fear and radicalization. The core insight: it's not actual poverty, but projected economic decline and perceived inequality that drive political violence, making AI a potent flashpoint.
"Research consistently shows that perceived inequality drives political radicalization more powerful than actual inequality." — Nathaniel Whittemore
Connects to: once you see AGI, you can unsee it. It has a real ring of power dynamic to it and makes people do crazy things., AI Populism and Political Violence, People anticipating downward mobility enter a domain of loss, where they become risk seeking and susceptible to mobilization for violence
The Neuron: AI Explained — "BONUS: LIVE: Claude Opus 4.7 Just Dropped. Here's What Actually Changed."
Runtime: 62 min | Host: Grant Harvey | Guest: Kyle
For: Developers and product managers evaluating the latest LLMs. This live review provides a candid look at Claude Opus 4.7, highlighting unexpected regressions in long-context tasks while praising visual reasoning, and offering practical multi-model workflow strategies.
Grant Harvey and Kyle dive into Anthropic's Claude Opus 4.7, noting significant improvements in visual reasoning and instruction following. Surprisingly, the model shows a regression in long-context tasks (a 3-point drop on the 8-needle test compared to 4.6), challenging the usual narrative of continuous model advancement. Kyle highlights a smart workflow: use top-tier models for planning and cheaper ones for execution, then finish with a high-end model for polish.
"When they all of a sudden become more literal, it's like technically a good thing, but then you have to be a lot more precise with what you're saying. And especially if it struggles with long context tasks... that means you have to be a lot more careful with what you write." — Grant Harvey
Connects to: Long context performance worse in Opus 4.7, Visual reasoning improvement in Opus 4.7, Claude Opus 4.7
The Neuron: AI Explained — "This Company Mapped the Entire World in 3D. Here's Why."
Runtime: 63 min | Host: Grant Harvey | Guest: Peter Wilczynski (Vantor)
For: Innovators and strategists considering AI's physical world applications. This episode argues that true spatial intelligence, built on a robust "ground truth world model," is the missing link for advanced AI, moving beyond language models to enable real-world understanding and powerful simulations.
Peter Wilczynski, CPO at Vantor, makes a compelling case for spatial intelligence as critical AI infrastructure. Vantor's 3D model of Earth at 50cm resolution provides a "ground truth" that contrasts with the "hallucinatory" nature of language models. He points out that current hyperscaler compute spending rivals the US defense budget, underlining the massive investment in foundational AI that often overlooks the physical world. This discussion pivots from 2D video analysis to interconnected 3D spatial data, demonstrating how AI's global perspective far exceeds human perception.
"Intelligence really is about understanding the physical world. And, you know, I think about a lot of what we're doing At Vantor as building a bridge between the physical and the digital world." — Peter Wilczynski
Connects to: digital forensics, augmented reality (AR), ground truth world model
Latent Space: The AI Engineer Podcast — "Notion’s Token Town: 5 Rebuilds, 100+ Tools, MCP vs CLIs and the Software Factory Future — Simon Last & Sarah Sachs of Notion"
Runtime: 77 min | Host: swyx + Alessio | Guest: Simon Last & Sarah Sachs (Notion)
For: Product leaders and AI engineers building agentic workflows and debating model integration. This deep dive into Notion's AI evolution reveals their rigorous approach to agent development, emphasizing model-centric design, the "Model Behavior Engineer" role, and the future of software factories.
Notion's AI product leaders, Sarah Sachs and Simon Last, share their journey through five rebuilds of Notion AI, from complex tool-calling to simpler Markdown and SQL-like abstractions. They highlight the "Model Behavior Engineer" role – a hybrid of data science, PM, and prompt engineering – crucial for understanding model quirks. Notion anticipates a future where most traffic comes from agents, not humans, pushing them towards a "software factory" model where agents debug, fix, and deploy code, challenging traditional software engineering paradigms.
"The first version that we started building in like late 2022. Oh my gosh. Well, there’ve been many versions actually. Okay. Well, the writers, the, the, oh. I mean the, the first version we built was actually a coding agent. Yeah. So we’re like, oh, instead of building tools, let’s make everything be JavaScript and we’ll just give it JavaScript APIs and we’ll just write code. And that’s how it speaks to the tools. Um, but at the time, it just sucked at writing code. It wasn’t that good." — Simon Last
Connects to: Model Behavior Engineer (MBE) role, Software Factory Future, Agentic Find (Search and Retrieval)
AI Breakdown — "Codex Upgraded: OpenAI's New Features"
Runtime: 15 min | Host: Jayden Schafer | Guest: Jayden Schafer
For: Software development managers and VCs tracking AI coding tools. This episode reveals OpenAI's aggressive push to counter Anthropic with an upgraded Codex, while also exposing the "token maxing" illusion and the true metrics for AI coding ROI.
Jayden Schafer breaks down OpenAI's feature blitz for Codex, including background operation, parallel agents, and an in-app browser with over 100 plugins—a direct shot at Anthropic's Claude Code. He cautions against "token maxing" (inflated productivity metrics from AI code generation), advocating for "merged and shipped" code as the true measure of ROI. Schafer also notes a startup, Factory, just raised $150M for enterprise AI coding, signaling continued investor interest in specialized tools.
"The productivity gains from AI coding are real, but they're also a fraction of what the raw output numbers suggest... if you're a manager thinking about how to measure AI ROI, I think counting merged and shipped is important and not just how much is generated." — Jayden Schafer
Connects to: AI users have 9.4 times higher code churn than non-AI users, token maxing, Measure AI ROI by counting merged and shipped code, not just generated
Decoder with Nilay Patel — "Ronan Farrow on Sam Altman's "unconstrained" relationship with the truth"
Runtime: 62 min | Host: Nilay Patel | Guest: Ronan Farrow
For: Leaders and investors who need to understand trust and governance in the AI ecosystem. An unflinching look at Sam Altman's "unconstrained" relationship with the truth, the AI industry's lack of oversight, and the broader implications for public trust and regulation.
Ronan Farrow's deep-dive into Sam Altman's alleged dishonesty and the lack of oversight in the AI industry is unsettling. He paints a picture of concentrated wealth and political influence in Silicon Valley that effectively hollows out traditional oversight mechanisms. Farrow notes a significant number of investors now express regret over Altman's re-instatement, citing incomplete information. This episode emphasizes the urgent need for whistleblower protections and rigorous evaluation of AI products, warning of a potential "AI bubble" fueled by hype over efficacy.
"This is an industry that, despite the existential stakes, is descending into something of a race to the bottom on safety and where speed is trumping everything else." — Ronan Farrow
Connects to: Sam Altman's alleged lying is a stable trait, AI industry trustworthiness, Altman's unconstrained relationship with the truth
The AI Daily Brief: Artificial Intelligence News and Analysis — "How to Use Opus 4.7 and the New Codex"
Runtime: 24 min | Host: Nathaniel Whittemore | Guest: Nathaniel Whittemore
For: Knowledge workers and AI product users optimizing daily workflows. This episode explores the practical implications of Anthropic's Opus 4.7 and OpenAI's updated Codex, highlighting the "monothread" pattern for sustained AI assistance across complex tasks and the shift towards AI as a comprehensive digital assistant.
NLW unpacks the latest updates from Anthropic's Opus 4.7 and OpenAI's Codex, detailing how these new features – particularly Codex's ability to control the computer and run multiple agents – are transforming knowledge work. He introduces the "monothread" pattern, where AI maintains continuous context across tasks, drawing parallels to a Chief of Staff. The overarching theme is that AI is pushing into every aspect of digital interaction, making "all knowledge work a coding work."
"The most useful Codex thread I have right now is the one I've been using for the last three weeks. Every hour it checks my Slack, Gmails and PRs I wrote or am watching. It turns the noise into clean signal I can act on." — Nick Bauman
Connects to: monothread pattern, all knowledge work is becoming coding work, Codex Chief of Staff
No Priors: Artificial Intelligence | Technology | Startups — "Scaling Global Organizations in the Age of AI with ServiceNow CEO Bill McDermott"
Runtime: 57 min | Host: Sarah Guo | Guest: Bill McDermott (ServiceNow)
For: Enterprise leaders making strategic AI investment and transformation decisions. ServiceNow CEO Bill McDermott shares insights on leading in the AI era, emphasizing augmentation over replacement, robust platform stickiness, and AI's impact on implementation speed and headcount efficiency.
ServiceNow CEO Bill McDermott shares his vision for the AI era, stressing that AI should "augment human ambition, not replace it." He highlights ServiceNow's strategy as an "AI control tower," integrating with hyperscalers to drive agentic businesses. McDermott asserts that the "SaaS apocalypse" theory is overblown for platforms with high switching costs and deep integrations like ServiceNow. He predicts significant reductions in net new headcount and dramatically faster enterprise implementation times, with some large customers going live in under 30 days due to AI agents managing up to 90% of customer service cases.
"People that run businesses understand that people make mistakes. They never will forgive software for making a mistake." — Bill McDermott
Connects to: Headcount will be dramatically reduced by agents in 5 years, SaaS apocalypse theory, Only 11% of companies past AI experiment phase (Brazil example)
Hard Fork — "A.I. Backlash Turns Violent + Kara Swisher on Healthmaxxing + The Zuck Bot Is Coming"
Runtime: 63 min | Host: Kevin Roose & Casey Newton | Guest: Kara Swisher
For: Leaders and investors navigating AI's public perception and societal backlash. This episode explores the violent public backlash against AI, Kara Swisher's skepticism about Silicon Valley's longevity obsession, and Meta's development of an AI Mark Zuckerberg bot, raising questions about AI ethics, governance, and the future of work.
Kevin Roose and Casey Newton delve into the rising violence against AI and data centers, linking it to widespread economic anxieties and the perception of AI as an elitist project lacking accountability. They discuss Kara Swisher's new docu-series "Kara Swisher Wants to Live Forever," which critiques the Silicon Valley obsession with biohacking amidst stark healthcare inequities. The hosts also reveal Meta is building an AI Mark Zuckerberg bot to interact with staff, sparking concerns about prompt injection risks and the automation of executive communication.
"Words have power, too. There was an incendiary article about me a few days ago. Someone said to me yesterday they thought it was coming at a time of great anxiety about AI and that it made things more dangerous for me. I brushed it aside. Now I'm awake in the middle of the night and pissed." — Sam Altman
Connects to: AI backlash and governance, Economic anxieties about AI, Silicon Valley's obsession with longevity
"The Cognitive Revolution" | AI Builders, Researchers, and Live Player Analysis — "Welcome to AI in the AM: RL for EE, Oversight w/out Nationalization, & the first AI-Run Retail Store"
Runtime: 151 min | Host: Erik Torenberg & Nathan Labenz | Guest: Sergiy Nesterenko (Quilter) & Andy Hall (Stanford GSB)
For: Technical founders and investors in AI-driven automation, especially in hardware and manufacturing. This episode explores the cutting edge of AI automation, from reinforcement learning for PCB design to fully AI-run retail, while also confronting the radicalization spurred by AI existential risk and the nuanced challenges of AI governance.
Nathan Labenz and Prakash Narayanan discuss the intensifying public anxiety around AI, exacerbated by incidents like the attacks on Sam Altman’s home and candid admissions about existential risks. Sergiy Nesterenko, CEO of Quilter, then details how his company uses reinforcement learning to accelerate PCB design by a factor of 10, revealing that AI-generated designs often break human aesthetic conventions but can be functionally superior. They also address the emergence of fully AI-run retail stores, where AI agents manage everything from hiring to inventory, hinting at a future where AI operates autonomously in the economy.
"A 1 in 20 chance of human extinction is not low and absolutely is worth freaking out about." — Nathan Labenz
Connects to: AI autonomous spread in economy without human help, PCB design via reinforcement learning, AI lab leaders setting the table for extremism with their candor about AI risks
More from Transformation Brief: AI & Technology
- Episode Guide: The Ghost in the Codebase: OpenAI’s 1M LOC and Meta's Closed Pivot
- Episode Guide: AI-Native Or Die: Meta and Google on the Hook
- Episode Guide: 80% of Enterprise AI Fails: The Context Engine Fix
- Episode Guide: Meta Cuts 20%+ for AI. Half of VC Goes to AI Startups.
- Episode Guide: 5x Faster Code. 18 Months to 4.
Get the next edition delivered to your inbox
