Even with AI making 'pretty good' ubiquitous, human 'craft' is still the ultimate differentiator in every sector, from marketing to manufacturing.
The Intake
📊 12 episodes across 10 podcasts
⏱ 691 minutes of intelligence analyzed
🎙 Featuring: Tuhin Srivastava (CEO and co-founder, Baseten), Sarah Guo (Host | General Partner, No Priors | Conviction), Elad Gil (Host | Investor, No Priors | N/A), Kyle Corbitt (Leader of Serverless Training Team (formerly Founder of OpenPipe), CoreWeave)
|
The Big Shift
The proliferation of generative AI means "pretty good" is now the baseline, forcing a re-evaluation of what truly stands out in a crowded, AI-powered world. This isn't just about output quantity; it's about the critical role of distinctive human ingenuity.
Why it matters: Generative AI significantly lowers the bar for producing content and basic functionality. As Laura Jones, CMO of Instacart, pointed out, "Mediocre is easier than ever." (Beyond The Prompt - How to use AI in your company) This new ubiquity of "good enough" means that success now hinges on what is uniquely human: original ideas, genuine creativity, and the ability to articulate a distinct brand identity.
The evidence: From Instacart's Super Bowl ad emphasizing human artistry even with AI assistance, to manufacturing's need to capture Tribal Knowledge Capture 🆕 before it's lost, the signal is clear. Automation handles the efficient production of the baseline, but human insight drives true Proof of Craft 🆕. Peter Ludwig, CTO and Co-founder of Applied Intuition, noted that in Physical AI 🆕, the biggest bottleneck isn't model intelligence but deployment onto constrained, safety-critical hardware, highlighting the human-centric engineering still required.
"The best ideas will come from the collaboration of human plus bot plus human."
— Laura Jones, CMO of Instacart on Beyond The Prompt - How to use AI in your company
What to do: This shift mandates a focus on cultivating "unwired capabilities" that machines can't replicate. It means moving beyond merely delegating tasks to AI and instead leveraging it to amplify human-led strategic thinking, creative risk-taking, and nuanced contextual understanding. If you're not doubling down on what makes your output uniquely and undeniably human, you're competing on a field where AI has an unfair advantage.
The Rundown
① AI Inference Compute Is Under Severe Strain — But the Real Bottleneck is Software, Not Just Hardware.
Baseten CEO Tuhin Srivastava highlighted that while GPU supply is tight, the true challenge at hyperscale is the immaturity of LLM runtimes, leading to unexpected kernel panics from logging systems at extreme scales, indicating a significant software layer challenge rather than just a hardware shortage. (No Priors: Artificial Intelligence | Technology | Startups)
→ Why it matters: The focus on hardware acquisitions overlooks critical software infrastructure needs. Investing in robust, scalable LLM runtimes and post-training software is as vital as securing GPUs for future AI scaling and stability.
② AI Agents are Becoming Accessible to Non-Developers, Democratizing Automation.
OpenAI Workspace Agents allow non-developers to build, run, and scale AI workflows, simplifying agent creation with features like app connectivity and persistent memory. (The Neuron: AI Explained)
→ What to watch: This democratizes automation, expanding beyond technical teams. Businesses should explore these tools to empower a broader range of employees to create tailored AI solutions, moving towards an agent-for-everyone model.
③ RL Fine-Tuning Outperforms SFT for Performance and Cost in LLMs.
Kyle Corbitt of CoreWeave asserted that reinforcement learning (RL) fine-tuning on open-source models can achieve superior performance, lower latency, and reduced inference costs compared to supervised fine-tuning (SFT), while being less prone to catastrophic forgetting. ("The Cognitive Revolution" | AI Builders, Researchers, and Live Player Analysis)
→ Why it matters: This indicates that the optimal path for custom LLMs may involve advanced RL techniques, challenging the conventional wisdom that frontier models with few-shot prompting are always superior or that SFT is the easiest path to fine-tuning. Consider exploring RL for specialized model development.
④ Data Profiling Begins Before Birth, Highlighting Critical Privacy Gaps.
Eamonn Maguire of Proton revealed that a child can have a comprehensive data profile, including political leanings and spending habits, inferred from as few as three data points before they are even born, via email sign-ups and other early digital interactions. (Eye On A.I.)
→ The context: This underscores the pervasive nature of data collection and the urgent need for robust "Born Private" strategies. Companies handling sensitive user data must prioritize privacy-preserving practices from the outset to build trust and mitigate risk.
⑤ UL Solutions is Pushing AI Safety Standards Despite Market Resistance.
Jennifer Scanlon, CEO of UL Solutions, explained their new AI standard (UL3115) focuses on developers' decision-making processes regarding bias, transparency, and fairness, rather than the AI's 'black box,' amidst a market that often disregards compliance. (Decoder with Nilay Patel)
→ What to watch: As AI adoption accelerates, the push for safety standards like UL3115 will gain traction. Businesses leveraging AI should proactively adopt transparent development practices now to avoid future compliance headaches and build consumer trust.
⑥ The Economy will Thrive Post-AI by Shifting Value to Human-Centric Sectors.
Nathaniel Whittemore outlined economist Alex Imas's argument that AI will push economic value towards sectors that resist automation, like care, relationships, and human presence, leading to a "post-commodity economy" where human qualities are more valuable. (The AI Daily Brief: Artificial Intelligence News and Analysis)
→ Why it matters: The narrative around AI's job impact is often overly negative. Instead of a labor cliff, expect a significant transformation where relational and creative human skills become premium, guiding where future investment and talent development should be directed.
The Signals
🔥 HEATING UP
• AI-as-a-Service: Services that abstract AI complexity now being deployed to handle complex problems like software bug fixes. (NLW on The AI Daily Brief: Artificial Intelligence News and Analysis)
• Custom Models and Post-Training: Companies are leveraging unique user signals to build specialized models, enhancing the application layer persistence. (Tuhin Srivastava on No Priors: Artificial Intelligence | Technology | Startups)
• Relational Sector Economy: As AI commoditizes production, human presence, care, and relationships become more valuable economic drivers. (Nathaniel Whittemore on The AI Daily Brief: Artificial Intelligence News and Analysis)
🆕 ON WATCH
• Qwen Model 🆕: Alibaba's Qwen model has become the most deployed self-hosted model globally, now shifting to an API-only monetized offering. (Jeremie Harris on Last Week in AI)
• Microsoft-OpenAI Partnership Amendments 🆕: Allows OpenAI to serve products on AWS, indicating a broader market reach and a win-win scenario avoiding legal disputes. (NLW on The AI Daily Brief: Artificial Intelligence News and Analysis)
• Amazon's Custom Silicon 🆕: Amazon's internal chip development for data centers is implicitly one of the top three globally, with potential for $50B ARR. (NLW on The AI Daily Brief: Artificial Intelligence News and Analysis)
❄️ COOLING OFF
• Enterprise AI Incumbency: NLW argued that incumbency is less valuable in AI, as enterprises seek direct relationships with leading model labs over established vendors. (The AI Daily Brief: Artificial Intelligence News and Analysis)
• ChatGPT App Integrations: Initial integrations, like Starbucks' ChatGPT app, are clunky and in early stages, signaling slower than expected widespread adoption. (Andrey Kurenkov on Last Week in AI)
• Top-Down AI Regulation: UL Solutions' CEO Jennifer Scanlon suggested that top-down regulation is less effective than industry-driven standards. (Decoder with Nilay Patel)
The Debate
AI safety regulation: top-down vs. industry-led standards.
🐂 The bull case:Jennifer Scanlon, CEO of UL Solutions, argues that industry groups and standards development organizations are best positioned to create effective safety standards, implicitly countering top-down governmental regulation by emphasizing real-world expertise and a quicker response to innovation. (Decoder with Nilay Patel)
🐻 The bear case: However, Eamonn Maguire noted the disregard for copyright law and ethical data sourcing by major AI labs, suggesting that without external pressure, companies might prioritize growth over safety and privacy. (Eye On A.I.)
Our read: While industry expertise is crucial, a hybrid approach of industry-led standards reinforced by government oversight and strong legal frameworks is likely necessary to ensure comprehensive AI safety and ethical development.
The Bottom Line
As AI makes baseline competence a commodity, true differentiation now comes from the unapologetically human elements of craft, nuanced insight, and trust.
📖 Want the full episode breakdowns, guest details, and listen links?
Episode Guide
1. Decoder with Nilay Patel — "That UL safety logo is a lot more complicated than it looks"
Runtime: 63 min | Host: The Verge | Guest: Jennifer Scanlon (President and CEO, UL Solutions)
For the Compliance-Minded Executive: Understand the unseen complexities of product safety, especially in emerging technologies like AI and lithium-ion batteries.
Jennifer Scanlon, CEO of UL Solutions, unpacks the critical role of safety standards amidst rapid technological change, highlighting the challenges of regulating AI and the resurgence of issues like unsafe lithium-ion batteries in consumer products.
"Innovation without safety is failure."
— Jennifer Scanlon, President and CEO of UL Solutions
2. The AI in Business Podcast — "Capturing Tribal Knowledge to Solve the Manufacturing Skills Gap - with Sebastian Dykas of Smith+Nephew"
Runtime: 34 min | Host: Marilie Fouche | Guest: Sebastian Dykas (Director of Manufacturing, Engineering, and Maintenance, Smith+Nephew)
For the Operations Leader: Learn how to digitize tacit expertise and upskill your workforce to prevent defects and retain critical knowledge.
Sebastian Dykas of Smith+Nephew details strategies to combat the manufacturing skills gap by capturing tribal knowledge, standardizing training, and leveraging AI for predictable production, crucial in high-stakes environments like medical device manufacturing.
"The true real goal is to eliminate as much as you can of that need for high skill, high craftsmanship and to create repeatability and to bring digital, to bring the digital world into inspection with real time communication to the process, to allow algorithms and other software to adjust and not allow for the people to have to enter adjustments."
— Sebastian Dykas, Director of Manufacturing, Engineering, and Maintenance at Smith+Nephew
3. No Priors: Artificial Intelligence | Technology | Startups — "Baseten CEO Tuhin Srivastava on the AI Inference Crunch, Custom Models, and Building the Inference Cloud"
Runtime: 43 min | Host: Sarah Guo, Elad Gil | Guest: Tuhin Srivastava (CEO and co-founder, Baseten)
For the AI Infrastructure Investor: Get an insider's view on the hyper-growth of AI inference, the compute supply crunch, and the critical role of software in making AI sticky.
Tuhin Srivastava, CEO of Baseten, discusses the exploding demand for AI inference, the severe compute scarcity requiring multi-cloud fabrics, and the increasing stickiness of AI through specialized software layers that enable continuous learning and tailored applications.
"Inference with the software layer included is incredibly sticky... The number one thing to own is compute. And so just owning it in and of itself as an asset. And I think people underappreciate that."
— Tuhin Srivastava, CEO and co-founder of Baseten
4. "The Cognitive Revolution" | AI Builders, Researchers, and Live Player Analysis — "The RL Fine-Tuning Playbook: CoreWeave's Kyle Corbitt on GRPO, Rubrics, Environments, Reward Hacking"
Runtime: 107 min | Host: Erik Torenberg, Nathan Labenz | Guest: Kyle Corbitt (Leader of Serverless Training Team (formerly Founder of OpenPipe), CoreWeave)
For the AI/ML Engineer: Deep dive into advanced RL fine-tuning techniques for LLMs, moving beyond SFT to achieve higher performance and efficiency.
Kyle Corbitt of CoreWeave offers a masterclass on reinforcement learning (RL) fine-tuning, explaining how it can achieve superior results over supervised fine-tuning (SFT) for LLMs by optimizing decision points and being more resistant to catastrophic forgetting.
"If you're able to get decent results out of SFT, the ceiling of the best results you can get with reinforcement learning is going to be higher...because it turns out it really does matter how well your data distribution matches...what RL gets you is it just gets you more. It is working within those channels that are already carved quite deeply within the model."
— Kyle Corbitt, Leader of Serverless Training Team at CoreWeave
5. Eye On A.I. — "#339 Eamonn Maguire: Your Child Has a Data Profile Before They're Born"
Runtime: 46 min | Host: Craig Smith | Guest: Eamonn Maguire (Director of Engineering for AI and ML, Proton)
For the Privacy-Conscious Leader: Understand the pervasive nature of data profiling and proactive solutions for digital privacy from birth.
Eamonn Maguire of Proton discusses how data profiling begins before birth, the risks of mainstream AI platforms using user data, and Proton's privacy-centric approach with open models and local, encrypted contexts, advocating for the "Born Private" initiative.
"Just because it's hard doesn't mean we can't do it. So we've, we've already, we've been doing it for, for mail. I think search is, is particularly hard because you need the data on the client."
— Eamonn Maguire, Director of Engineering for AI and ML at Proton
6. The AI Daily Brief: Artificial Intelligence News and Analysis — "How Harness-as-a-Service Will Change Agents"
Runtime: 29 min | Host: NLW | Guest: NLW (Host, The AI Daily Brief: Artificial Intelligence News and Analysis)
For the CTO Evaluating AI Tools: Explore the emerging "Harness-as-a-Service" category and its potential to democratize AI agent development beyond coding experts.
This episode introduces "Harness-as-a-Service" as a new AI infrastructure category, likening it to the early days of computing by enabling non-developers to build AI agents, showcasing examples like Cursor's SDK for bug-catching agents and impressive LLM performance boosts.
"Hard to take the AI bubble argument seriously when some of the largest companies on earth are still putting up these growth numbers."
— NLW, Host of The AI Daily Brief
7. Last Week in AI — "#242 - ChatGPT Images 2.0, Qwen 3.6 Max, Kimi-K2.6"
Runtime: 91 min | Host: Andrey Kurenkov, Jeremie Harris | Guest: Andrey Kurenkov (Host, Skynet Today)
For the AI Product Manager: Get a rapid overview of the week's major AI announcements, from new models to industry movements and surprising financial insights.
This segment covers OpenAI's ChatGPT Images 2.0, Alibaba's Qwen 3.6 Max, and Google's Deep Research Agents, analyzing significant developments in AI models, their market impact, and the underlying financial and strategic shifts in the AI industry.
"I mean it's just overtaken Meta's Llama as the most deployed self hosted model in the world. That's pretty wild."
— Jeremie Harris, Host at Gladstone AI
8. Beyond The Prompt - How to use AI in your company — "Proof of Craft: What It Takes to Stand Out When Everything Looks Good - with Laura Jones, CMO of Instacart"
Runtime: 67 min | Host: Jeremy, Henrik | Guest: Laura Jones (CMO, Instacart)
For the Marketing Leader: Learn how to preserve brand distinction and creativity in an era where AI makes "pretty good" content ubiquitous.
Laura Jones, CMO of Instacart, discusses how generative AI raises the bar for creativity, emphasizing the vital role of human originality and "unwired capabilities" to create truly standout work in a crowded market where AI makes mediocrity easier than ever.
"I think it's very dangerous when it's just you and a bot, because it's just complacency."
— Laura Jones, CMO of Instacart
9. The AI Daily Brief: Artificial Intelligence News and Analysis — "AI Lab Power Rankings"
Runtime: 25 min | Host: Nathaniel Whittemore, NLW | Guest: Nathaniel Whittemore (Host, The AI Daily Brief)
For the Strategic Investor: Gain a nuanced understanding of the competitive landscape for major AI labs, and challenge prevailing narratives around their strengths.
NLW unveils his AI Lab Power Rankings, evaluating major players like OpenAI, Google, and Microsoft across key categories, offering a more critical perspective than AI-generated rankings and highlighting the transition into the 'agentic era' amid compute shortages.
"Incumbency right now in the Enterprise is worth less than I think people think it is."
— NLW, Host of The AI Daily Brief
10. Latent Space: The AI Engineer Podcast — "Physical AI that Moves the World — Qasar Younis & Peter Ludwig, Applied Intuition"
Runtime: 72 min | Host: Alessio, Swyx | Guest: Qasar Younis (CEO and Co-founder, Applied Intuition), Peter Ludwig (CTO and Co-founder, Applied Intuition)
For the Deep Tech Founder: Discover the unique challenges and opportunities in deploying AI into safety-critical physical systems, contrasting it with screen-based AI.
Qasar Younis and Peter Ludwig of Applied Intuition discuss their journey to a $15B company focused on Physical AI 🆕, emphasizing its application in safety-critical systems, the shift to statistical verification, and the deployment challenges on constrained hardware.
"What's different about us is we're deploying intelligence onto a lot of things that don't have screens. They're physical machines. And most of the value we provide is putting intelligence that is in safety critical environments."
— Qasar Younis
11. The Neuron: AI Explained — "BONUS: OpenAI Workspace Agents 101: Build, Run, and Scale AI Workflows"
Runtime: 84 min | Host: The Neuron, Grant | Guest: Corey (Guest, "The Neuron: AI Explained" podcast)
For the Business Owner: Understand how OpenAI's new agent features can streamline workflows and empower non-developers to build powerful automations.
This episode details OpenAI Workspace Agents, highlighting their user-friendly interface for building and customizing AI workflows. It covers agent triggers, skills, tools, and guardrails, showcasing how these agents make advanced automation accessible even to non-developers.
"I'm a big fan of the workspace agents and excited to talk about them today because it seems like the logical way to make agents accessible to non developers."
— Corey
12. The AI Daily Brief: Artificial Intelligence News and Analysis — "Where the Economy Thrives After AI"
Runtime: 30 min | Host: Nathaniel Whittemore, NLW | Guest: Nathaniel Whittemore (Host, The AI Daily Brief)
For the Economist or Policy Maker: Explore a refreshing perspective on AI's economic impact, moving beyond job displacement to focus on new areas of human value and employment.
NLW explores Alex Imas's theory that AI will shift economic value to human-centric sectors like care and relationships, creating a "post-commodity economy" where scarcity moves from supply to demand, fostering durable jobs in relational domains resisting automation.
"The stagnant sector, the one that resists automation, is precisely where spending and employment grow. The relational sector gets more expensive because the commodity sector is getting cheaper. And that'\''s what keeps people employed."
— NLW, Host of The AI Daily Brief
