The AI Fundamentalists

As multi-step agentic AI systems evolve, performance is increasingly driven by orchestration harnesses and stepwise outcome verification rather than raw model scale. While gated sub-agent architectures help prevent error cascades, the industry faces a sharp reckoning around vibe coding security vulnerabilities and unsustainable token costs. Paradoxically, generic AI travel planning tools are causing widespread itinerary homogenization, which in turn is driving up the market value and prestige of true human expertise.

• Agentic Harnesses & Stepwise Verification: Sub-agent harnesses and gated completions evaluate individual steps to eliminate error cascades, maintaining ~90% accuracy on multi-step tasks.
• The "Vibe Coding" Security Vulnerability: Unreviewed AI-generated code accelerates security flaws, as cataloged in Georgia Tech's Vibe Security Radar.
• Automated Exploitation Flywheel: AI models act as automated "script kiddies," speeding up the arms race by rapidly exploiting known vulnerabilities rather than creating novel zero-days.
• Token Economics & High Compute Costs: Escalating token expenses and infrastructure debt have led companies to move away from "token maxing," with human labor sometimes proving more cost-effective.
• Harnesses Over Brute-Force Scaling: Combining smaller or open-weight models with tailored harnesses, skills, and Model Context Protocols (MCPs) often delivers superior economic ROI compared to massive LLMs.
• First-Principles Governance Beyond "Kill Switches": Effective agentic governance requires fundamental system design, prompt edit tracking, and objective review rather than relying solely on reactive kill switches.
• The "Travel Agent 2.0" Homogenization Paradox: Generic AI travel tools generate impractical, 16-hour itineraries lacking local context, which unexpectedly elevates the prestige and necessity of real human travel agents.
• Right-Sizing Tools & Deterministic Systems: Traditional rule-based expert systems and optimization algorithms remain far more token-efficient and accurate for nuanced preference matching than brute-force LLMs.
• The "Cyborg" Collaboration Model: Human-AI hybrid workflows consistently outperform pure AI autonomy by pairing algorithmic efficiency with essential human context and oversight.

Mentioned in the episode: Vibe Security Radar (vibesecradar.com/) and GitHub on Vibe Security Radar (github.com/HQ1995/vibe-security-radar).

Previous related episodes: 

• Mechanism design: Building smarter AI agents from the fundamentals, part 1 
• Utility functions: Building smarter AI agents from the fundamentals, part 2 
• Linear programming: Building smarter AI agents from the fundamentals, part 3

What did you think? Let us know.

Do you have a question or a discussion topic for the AI Fundamentalists? Connect with them to comment on your favorite topics:

• LinkedIn - Episode summaries, shares of cited articles, and more.
• YouTube - Was it something that we said? Good. Share your favorite quotes.
• Visit our page - see past episodes and submit your feedback! It continues to inspire future episodes.

Andrew and Sid break down the hidden costs of AI tokens, why current prices are artificially low, and whether AI tokens could become the next global commodity.

As AI adoption surges and agentic workflows burn through compute, the underlying economics of large language models are reaching a critical inflection point. Join us to explore the rapidly shifting landscape of "tokenomics," the staggering hardware constraints behind the scenes, and what the true market clearing price for AI might actually look like.

To help us unpack this, the hosts dive into the downstream effects of "token maxing," why true economic equilibrium in AI is far off, and how historical technological shifts like electricity can predict our AI future.

• Defining what a token actually is and how text is chunked and processed by specific models.
• The illusion of current token pricing and why heavy subsidization by tech giants obscures the true cost of production.
• Exploring the flawed "token maxing" trend and why organizations are improperly prioritizing raw AI usage over actual return on investment.
• The severe hardware constraints and geopolitical pressures, including skyrocketing GPU and RAM costs, that make running local infrastructure incredibly difficult.
• Analyzing the criteria for money to see if AI tokens can become a true currency, or if they are destined to act as a tradable commodity like oil.
• The "Jevons Paradox" of AI efficiency and why cheaper compute actually leads to massively increased, rather than decreased, usage.
• How the future of work will rely on "cyborging"—combining human talent with AI—to increase productivity, using the surprising resurgence of human travel agents as an example.

This episode is full of economic insights and forward-looking predictions that are sure to change how you think about your next API bill. As we move into a new era of AI, it's the perfect time to explore the fundamentals of the next frontier!

What did you think? Let us know.

Do you have a question or a discussion topic for the AI Fundamentalists? Connect with them to comment on your favorite topics:

• LinkedIn - Episode summaries, shares of cited articles, and more.
• YouTube - Was it something that we said? Good. Share your favorite quotes.
• Visit our page - see past episodes and submit your feedback! It continues to inspire future episodes.

In this episode of The AI Fundamentalists, hosts Andrew and Sid are joined by AI alignment and safety researcher Dr. Jillian Fisher to unpack the complex realities of political bias in Large Language Models. Dr. Fisher explains that bias isn't just a byproduct of noisy training data; it is also embedded directly into the architectural choices of the models, such as relying on a "majority vote" mechanism to determine the right answer.

The conversation explores why achieving true political neutrality in AI is widely considered impossible due to the inescapable human element involved in AI development. Instead, developers must rely on imperfect approximations of neutrality. Dr. Fisher breaks down approaches like "reasonable pluralism"—which attempts to present all reasonable sides of an argument—and flat-out refusal to answer, noting that both strategies come with distinct trade-offs for user agency and safety.

Listeners will also discover fascinating insights into the psychology of AI persuasion. Dr. Fisher highlights research showing that unlike humans, who typically persuade through empathy and storytelling, AI is most convincing to users through "information packing". Delivering dense walls of facts, combined with natural conversational fluency, can trick our brains into viewing the model as an unquestionable authority. Finally, the group discusses the critical need for socio-technical AI literacy, exploring how teaching the public about AI's limitations and its reliance on flawed internet data could be the ultimate tool for inoculating users against sycophantic behaviors and unwanted persuasion.

What did you think? Let us know.

Do you have a question or a discussion topic for the AI Fundamentalists? Connect with them to comment on your favorite topics:

• LinkedIn - Episode summaries, shares of cited articles, and more.
• YouTube - Was it something that we said? Good. Share your favorite quotes.
• Visit our page - see past episodes and submit your feedback! It continues to inspire future episodes.

In this episode we conclude our  series about Metaphysics and modern AI, we explore the definitions of consciousness, reasoning, and thinking to understand if AI possesses these traits. From examining legal accountability and the concept of personhood to analyzing human cognitive frameworks, we map out the differences between actual contemplative problem-solving and probabilistic pattern recognition. The episode covers: 

• Defining consciousness, reasoning, and what it means to be a "thinking thing"
• The Turing Test as a low bar and why natural language capabilities create the illusion of intelligence
• Accountability and agency: Why AI models like Claude are not legally recognized as persons
• Daniel Kahneman’s System 1 (fast heuristics) vs. System 2 (contemplative reasoning) thinking
• Why LLMs function primarily as System 1 pattern recognizers rather than true reasoners
• Complex systems, Descartes' dualism, and whether thinking is an emergent property requiring a physical body
• How chatbots use psychological mirroring, filler words, and pauses to trick human biases
• The dangers of anthropomorphizing AI driven by fear of change or financial incentives

This is the final episode in our metaphysics and AI series. You can find the previous episodes here:

• Metaphysics and modern AI: What is causality? 
• Metaphysics and modern AI: What is reality? 
• Metaphysics and modern AI: What is thinking? - Series Intro 

What did you think? Let us know.

Do you have a question or a discussion topic for the AI Fundamentalists? Connect with them to comment on your favorite topics:

• LinkedIn - Episode summaries, shares of cited articles, and more.
• YouTube - Was it something that we said? Good. Share your favorite quotes.
• Visit our page - see past episodes and submit your feedback! It continues to inspire future episodes.

As the AI landscape evolves, the methods we use to process structured data are undergoing a silent revolution. Join us to explore how Tabular Foundation Models (TFMs) are challenging the decade-long reign of tree-based algorithms, why the traditional "train and predict" workflow is being replaced by "in-context learning," and what this shift means for the future of resilient modeling.

To help us, Christoph Molnar, renowned expert in machine learning interpretability and author of the Mindful Modeler newsletter, joins us to share his perspective on the emergence of tabular transformers, the surprising power of synthetic data, and how to maintain model safety in a world without parameter updates.

• The decline of the "fit and predict" paradigm in tabular data
• Transformer architectures vs. traditional models like XGBoost and LightGBM
• In-context learning: Predicting without traditional training steps
• The role of Structural Causal Models (SCMs) in generating training data
• Why models trained on "math and probability" succeed on real-world datasets
• Hardware accessibility and running foundation models on local MacBooks
• Integrating SHAP values and conformal prediction for model interpretability
• The future of the data science workflow: One tool among many or a total shift?

This episode is full of technical insights and forward-looking predictions that are sure to change how you approach your next dataset. As we move into a new era of AI, it’s the perfect time to explore the fundamentals of the next frontier!

What did you think? Let us know.

Do you have a question or a discussion topic for the AI Fundamentalists? Connect with them to comment on your favorite topics:

• LinkedIn - Episode summaries, shares of cited articles, and more.
• YouTube - Was it something that we said? Good. Share your favorite quotes.
• Visit our page - see past episodes and submit your feedback! It continues to inspire future episodes.

As information sources have become abundant and attention spans have shortened in the age of AI, we take on the lost art of reading. Join us to explore why reading rates are falling, how that shift affects judgment and opportunity, and how interdisciplinary books help us see patterns across history, economics, and technology. 

To help us, Alisa Rusanoff, CEO of Eltech AI, joins us to share her perspective on reading, debate volume versus depth, and offer practical ways to reclaim attention and read with intention.

• Evidence on declining reading rates among adults, teens and children
• Noise versus signal in the attention economy
• Mental models and interdisciplinary synthesis for better decisions
• AI’s limits and why human integration still matters
• Cycles in debt, trade, demography, and geopolitics
• Fiction as a cultural sensor for lived experience
• Wealth gaps, polarization and the need for critical thinking
• Practical habits to train feeds and protect reading time
• Challenge to read, reflect, and apply insights

For people worried if they are reading enough:

• Reading just 1 book a year puts you in the top 60% of readers
• Read 4 books a year to be in the top 50% of readers
• Read 10 books a year to be in the top 20% of readers
• For those looking to be in the top 5% of readers, expect to read at least 50 books

This episode is full of research and fun connections that are sure to make you think positively about your commitment to reading. At the time of this episode, it's not too late to join the top 20% in 2026!

What did you think? Let us know.

Do you have a question or a discussion topic for the AI Fundamentalists? Connect with them to comment on your favorite topics:

• LinkedIn - Episode summaries, shares of cited articles, and more.
• YouTube - Was it something that we said? Good. Share your favorite quotes.
• Visit our page - see past episodes and submit your feedback! It continues to inspire future episodes.

In this episode of our series about Metaphysics and modern AI, we break causality down to first principles and explain how to tell factual mechanisms from convincing correlations. From gold-standard Randomized Control Trials (RCT) to natural experiments and counterfactuals, we map the tools that build trustworthy models and safer AI.

• Defining causes, effects, and common causal structures
• Gestalt theory: Why correlation misleads and how pattern-seeking tricks us
• Statistical association vs causal explanation
• RCTs and why randomization matters
• Natural experiments as ethical, scalable alternatives
• Judea Pearl’s do-calculus, counterfactuals, and first-principles models
• Limits of causality, sample size, and inference
• Building resilient AI with causal grounding and governance
  
  

This is the fourth episode in our metaphysics series. Each topic in the series is leading to the fundamental question, "Should AI try to think?"

Check out previous episodes:

• Series Intro
• What is reality?
• What is space and time?

If conversations like this sharpen your curiosity and help you think more clearly about complex systems, then step away from your keyboard and enjoy this journey with us.

What did you think? Let us know.

Do you have a question or a discussion topic for the AI Fundamentalists? Connect with them to comment on your favorite topics:

• LinkedIn - Episode summaries, shares of cited articles, and more.
• YouTube - Was it something that we said? Good. Share your favorite quotes.
• Visit our page - see past episodes and submit your feedback! It continues to inspire future episodes.

In this episode, Dr. Sebastian (Seb) Benthall joins us to discuss research from his and Andrew's paper entitled “Validity Is What You Need” for agentic AI that actually works in the real world. 

Our discussion connects systems engineering, mechanism design, and requirements to multi‑step AI that creates enterprise impact to achieve measurable outcomes.

• Defining agentic AI beyond LLM hype
• Limits of scale and the need for multi‑step control
• Tool use, compounding errors, and guardrails
• Systems engineering patterns for AI reliability
• Principal–agent framing for governance
• Mechanism design for multi‑stakeholder alignment
• Requirements engineering as the crux of validity
• Hybrid stacks: LLM interface, deterministic solvers
• Regression testing through model swaps and drift
• Moving from universal copilots to fit‑for‑purpose agents

You can also catch more of Seb's research on our podcast. Tune in to Contextual integrity and differential privacy: Theory versus application.

What did you think? Let us know.

Do you have a question or a discussion topic for the AI Fundamentalists? Connect with them to comment on your favorite topics:

• LinkedIn - Episode summaries, shares of cited articles, and more.
• YouTube - Was it something that we said? Good. Share your favorite quotes.
• Visit our page - see past episodes and submit your feedback! It continues to inspire future episodes.

Here it is! We review the year where scaling large AI models hit its ceiling, Google reclaimed momentum with efficient vertical integration, and the market shifted from hype to viability. 

Join us as we talk about why human-in-the-loop is failing, why generative AI agents validating other agents compounds errors, and how small expert data quietly beat the big models.

• Google’s resurgence with Gemini 3.0 and TPU-driven efficiency
• Monetization pressures and ads in co-pilot assistants
• Diminishing returns from LLM scaling
• Human-in-the-loop pitfalls and incentives
• Agents vs validation and compounding error
• Small, high-quality data outperforming synthetic
• Expert systems, causality, and interpretability
• Research trends return toward statistical rigor
• 2026 outlook for ROI, governance, and trust

We remain focused on the responsible use of AI. And while the market continues to adjust expectations for return on investment from AI, we're excited to see companies exploring "return on purpose" as the new foray into transformative AI systems for their business. 

What are you excited about for AI in 2026? 

What did you think? Let us know.

Do you have a question or a discussion topic for the AI Fundamentalists? Connect with them to comment on your favorite topics:

• LinkedIn - Episode summaries, shares of cited articles, and more.
• YouTube - Was it something that we said? Good. Share your favorite quotes.
• Visit our page - see past episodes and submit your feedback! It continues to inspire future episodes.

In this episode, we look at the actuarial principles that make models safer: parallel modeling, small data with provenance, and real-time human supervision. To help us, long-time insurtech and startup advisor David Sandberg, FSA, MAAA, CERA, joins us to share more about his actuarial expertise in data management and AI.

We also challenge the hype around AI by reframing it as a prediction machine and putting human judgment at the beginning, middle, and end. By the end, you might think about “human-in-the-loop” in a whole new way.

• Actuarial valuation debates and why parallel models win
• AI’s real value: enhance and accelerate the growth of human capital
• Transparency, accountability, and enforceable standards
• Prediction versus decision and learning from actual-to-expected
• Small data as interpretable, traceable fuel for insight
• Drift, regime shifts, and limits of regression and LLMs
• Mapping decisions, setting risk appetite, and enterprise risk management (ERM) for AI
• Where humans belong: the beginning, middle, and end of the system
• Agentic AI complexity versus validated end-to-end systems
• Training judgment with tools that force critique and citation

Cultural references:

• Foundation, AppleTV
• The Feeling of Power, Isaac Asimov
• Player Piano, Kurt Vonnegut

For more information, see Actuarial and data science: Bridging the gap.

What did you think? Let us know.

Do you have a question or a discussion topic for the AI Fundamentalists? Connect with them to comment on your favorite topics:

• LinkedIn - Episode summaries, shares of cited articles, and more.
• YouTube - Was it something that we said? Good. Share your favorite quotes.
• Visit our page - see past episodes and submit your feedback! It continues to inspire future episodes.