Two Minds, One Model

Why orchestrate multiple AI agents when a single strong model is so capable? Jon walks through three distinct rationales — capability, parallel context, and trust — and uses Anthropic's Claude Mythos Preview and Project Glasswing as the live, industrial-scale case study.

Credits

Cover Art by Brianna Williams

TMOM Intro Music by Danny Meza

A special thank you to these talented artists for their contributions to the show.

Links and Reference

• Stanford 2026 AI Index Report: https://hai.stanford.edu/ai-index/2026-ai-index-report

• Claude Opus 4.7 announcement: https://www.anthropic.com/news/claude-opus-4-7

• Project Glasswing announcement: https://www.anthropic.com/glasswing

• Claude Mythos Preview — Frontier Red Team write-up: https://red.anthropic.com/2026/mythos-preview/

• Claude Mythos Preview — Alignment Risk Update: https://anthropic.com/claude-mythos-preview-risk-report

• Andon Labs Vending-Bench (the eval Jon describes): https://andonlabs.com/evals/vending-bench

• Mixture-of-Agents (Wang et al., June 2024): https://arxiv.org/abs/2406.04692

• Self-MoA / "Rethinking Mixture-of-Agents" (Lee et al., Feb 2025): https://arxiv.org (search by title)

• AI Control: Improving Safety Despite Intentional Subversion (Greenblatt et al., Dec 2023, Redwood Research): https://arxiv.org/abs/2312.06942

• Anthropic multi-agent research system blog: https://www.anthropic.com/engineering/built-multi-agent-research-system

• MAGDI — distilling multi-agent debate (Chen et al., early 2024): https://arxiv.org/abs/2402.01620

• MACA — Multi-Agent Consensus Alignment (Sept 2025): https://arxiv.org (search by title)

• Agent Arc — distilling multi-agent intelligence into a single LLM agent (Feb 2026): https://arxiv.org (search by title)

• Condorcet Jury Theorem (1785): https://plato.stanford.edu/entries/jury-theorems/

Abandoned Episode Titles

How to Build God and Then Email Yourself About It from the Park

Four PhDs and a Guy Who Thinks the Colosseum Invented Pasta

Mythos Cleaned Its Git History So You Wouldn't Have To

OpenBSD Spent 27 Years Hardening the Wrong Things

This episode deconstructs how production AI agents are actually built, introducing a six-component architecture framework (system prompt, model, tools, memory, orchestration loop, and execution environment) and comparing how Claude Code, Codex, OpenClaw, and Manus make fundamentally different trade-offs around local vs. cloud execution, autonomy vs. human oversight, and open source vs. commercial control. The hosts examine why coding agents matured first, why general-purpose agents face the unsolved "lethal trifecta" of security risks, and where the industry is converging on universal patterns while still making divergent bets.

Credits

Cover Art by Brianna Williams

TMOM Intro Music by Danny Meza

A special thank you to these talented artists for their contributions to the show.

Links and Reference

• Meta Muse Spark announcement: https://ai.meta.com/blog/introducing-muse-spark-msl/

• Anthropic Project Glasswing / Claude Mythos: https://www.anthropic.com/glasswing

•  Anthropic Mythos Preview technical details: https://red.anthropic.com/2026/mythos-preview/

• Google TurboQuant (ICLR 2026): https://research.google/blog/turboquant-redefining-ai-efficiency-with-extreme-compression/

• Let's Verify Step by Step (Lightman et al.): https://arxiv.org/abs/2305.20050

• METR Time Horizons: https://metr.org/time-horizons/

• METR: Measuring AI Ability to Complete Long Tasks: https://arxiv.org/abs/2503.14499

• Simon Willison's blog: https://simonwillison.net/

• Simon Willison: The Lethal Trifecta: https://simonwillison.net/2025/Jun/16/the-lethal-trifecta/

• OpenClaw (GitHub): https://github.com/OpenClaw/OpenClaw

• Peter Steinberger: OpenClaw, OpenAI and the future: https://steipete.me/posts/2026/openclaw

• Manus joins Meta: https://manus.im/blog/manus-joins-meta-for-next-era-of-innovation

• CrowdStrike on Mythos / Project Glasswing: https://www.crowdstrike.com/en-us/blog/crowdstrike-founding-member-anthropic-mythos-frontier-model-to-secure-ai/

• Model Context Protocol (MCP): https://modelcontextprotocol.io/

• Stuart Russell, Human Compatible (2019): https://www.penguinrandomhouse.com/books/566677/human-compatible-by-stuart-russell/

Abandoned Episode Titles

My Torn Hoodie Is Perfectly Fine, Thank You Very Much

ChatGPT Bought This Outfit for Me

The Lobster, the Sandbox, and the Wardrobe

It's Agents All the Way Down

This episode traces AI reasoning from human-designed external scaffolding (process reward models, test-time compute scaling) to internally emergent capability, culminating in DeepSeek R1's finding that a model rewarded only for correctness spontaneously learns to reason, self-correct, and backtrack without any explicit instruction to do so.

Credits

Cover Art by Brianna Williams

TMOM Intro Music by Danny Meza

A special thank you to these talented artists for their contributions to the show.

Links and Reference

• US appeals court fined lawyers https://www.sixthcircuitappellateblog.com/recent-cases/sixth-circuit-sanctions-attorneys-for-fake-citations-what-does-this-mean-for-use-of-ai/https://www.jdsupra.com/legalnews/the-ai-sanction-wave-145k-in-q1-1240943/#:~:text=In%20Whiting%20v.%20City%20of,cases%20presenting%20the%20same%20problems.

• CEO Krafton used ChatGPT to nullify $250M contract https://legaltalknetwork.com/podcasts/heels-in-the-courtroom/2026/04/ep-1006-when-clients-use-ai-the-new-risks-to-privilege-and-discovery/#:~:text=So%20the%20allegations%20were%20that,let%20ChatGPT%20be%20his%20lawyer.

• "Let's Verify Step by Step" https://arxiv.org/abs/2305.20050

• PRM800K dataset — 800,000 step-level human feedback labels, open-sourcedhttps://github.com/openai/prm800k

• Snell et al. paper on test-time compute scaling, published Aug 2024https://arxiv.org/abs/2408.03314

• "Chinchilla optimal" — paper on optimal scaling of parameters vs. datahttps://arxiv.org/pdf/2203.15556

• LangChain documented convergence in open SWE frameworkhttps://blog.langchain.com/open-swe-an-open-source-framework-for-internal-coding-agents/

• "Thinking Fast and Slow" by Kahneman, Dhttps://psycnet.apa.org/record/2011-26535-000

• T3 Code — Theo's Claude Code harness replacementhttps://www.youtube.com/watch?v=-7akxGb-lAM#:~:text=Theo%20Did%20It.,Gemini%20without%20the%20lock%2Din.

• DeepSeek R1 technical report, January 2025 https://arxiv.org/abs/2501.12948

• Uncanny Valley concepthttps://web.ics.purdue.edu/~drkelly/MoriTheUncannyValley1970.pdf

Abandoned Episode Titles

The Episode That Definitely Didn't Anthropomorphize Anything

Pump Harder: A Metaphor That Should Have Died But Absolutely Didn't

"Wait, Wait, Wait, Don’t Tell Me"

The One Where the Math Problem Checks Its Own Work and We All Get a Little Creeped Out

This episode is a practical guide to working with AI agents — covering what makes them different from chatbots, how to craft effective agentic prompts, how to calibrate trust and supervision across the autonomy spectrum, and best practices for coding, research, and personal assistant agents. John frames the core skill as delegation, not querying, and walks through the pitfalls that trip up new agent users.

Credits

Cover Art by Brianna Williams

TMOM Intro Music by Danny Meza

A special thank you to these talented artists for their contributions to the show.

Corrections: 

• When discussing an article on the “lethal trifecta,” John mistakenly names the author as Sam Willison, which is incorrect. The author’s correct name is Simon Willison. Our apologies for misspeak. Link to paper in the reference section.

• John incorrectly quoted the Manus tagline as “The AI that actually does things.” The original tagline was “The AI that DOES.” https://medium.com/@okkark.pro/from-shell-product-to-2-billion-the-manus-ai-story-nobody-saw-coming-d8308b57a42e

Links and Reference

• QuitGPT movement / 2.5M users / Pentagon deal: https://techcrunch.com/2026/03/02/chatgpt-uninstalls-surged-by-295-after-dod-deal/

• GPT 5.4 rush release: https://techcrunch.com/2026/03/05/openai-launches-gpt-5-4-with-pro-and-thinking-versions/

• Yann LeCun / AMI Labs / $1B seed round: https://techcrunch.com/2026/03/09/yann-lecuns-ami-labs-raises-1-03-billion-to-build-world-models/

• DeepSeek V4 / Huawei Ascend chips: https://medium.com/@michael_68282/share-deepseek-v4-will-come-only-when-deepseek-is-ready-for-it-to-come-not-before-a778e55c2655

• "Hunter Alpha" — mystery 1T parameter model on Open Router: https://medium.com/@him2696/the-mystery-of-hunter-alpha-the-anonymous-1-trillion-parameter-ai-taking-over-openrouter-9e4e94dc0cb8

• MIT deep learning heart failure prediction: https://news.mit.edu/2026/can-ai-help-predict-which-heart-failure-patients-will-worsen-0312

• Claude found 500+ zero-day vulnerabilities in Firefox: https://www.anthropic.com/news/mozilla-firefox-security

• Claude Code hitting $1B run rate: Jhttps://www.linkedin.com/posts/chintanzalani_claude-code-has-hit-1b-run-rate-revenue-activity-7402079378714923008-TtUp/

• Anthropic Cowork launch / January 2026: https://www.anthropic.com/webinars/future-of-ai-at-work-introducing-cowork

• METR time horizons for AI agents: https://metr.org/time-horizons/

• Anthropic "Eight Trends" blog post / 60% AI use / <20% delegation: https://claude.com/blog/eight-trends-defining-how-software-gets-built-in-2026

• Opus 4.6 release blog safety warnings: https://www.anthropic.com/news/claude-opus-4-6

• Simon Willison’s lethal trifecta: https://simonwillison.net/2025/Jun/16/the-lethal-trifecta/

• CrowdStrike report on OpenClaw: https://www.crowdstrike.com/en-us/blog/what-security-teams-need-to-know-about-openclaw-ai-super-agent/

• Next episode papers — RLVR and "Let's Verify Step by Step": https://arxiv.org/abs/2305.20050

Abandoned Episode Titles

"Penny Wise, Prompt Foolish"

"My Agent Exposed my API Key to the Internet and All I Got Was This Lousy Podcast Episode"

"Twenty Percent of the Time, It Ignores Me All the Time"

“Read the Diff, Not the Vibes”

This episode traces how prompt engineering evolved from informal tricks (tipping, role-playing, "take a deep breath") into three structured reasoning frameworks — Chain of Thought, Self-Consistency, and Tree of Thoughts — that dramatically improved LLM performance without changing the models themselves, culminating in the insight that intelligence in these systems is a latent resource unlocked by better scaffolding, not better weights.

Credits

Cover Art by Brianna Williams

TMOM Intro Music by Danny Meza

A special thank you to these talented artists for their contributions to the show.

Links and Reference

• Chain of Thought Prompting: Wei, J., Wang, X., Schuurmans, D., et al. (2022). "Chain-of-Thought Prompting ElicitsReasoning in Large Language Models." NeurIPS 2022. arXiv: 2201.11903

• Self-Consistency: Wang, X., Wei, J., Schuurmans, D., et al. (2022). "Self-Consistency Improves Chain of Thought Reasoning in Language Models." ICLR 2023. arXiv: 2203.11171

• Tree of Thoughts: Yao, S., Yu, D., Zhao, J., et al. (2023). "Tree of Thoughts: Deliberate Problem Solving with Large Language Models." NeurIPS 2023. arXiv: 2305.10601

• "Take a deep breath and think carefully" improves performance:: Yang, C., Wang, X., Lu, Y., et al. (2023). "Large Language Models as Optimizers." arXiv:2309.03409. 

• Christmas / holiday performance degradation caveat: This claim was popularized on social media and discussed on platforms like X/Twitter and Hacker News in late 2023. A blog post by Rob Lynch (December 2023) ran some informal tests. No peer-reviewed

• study has definitively confirmed this effect. Consider adding a caveat.

• Cleverbot:: Cleverbot (1997–2023). Originally created by Rollo Carpenter. Website: cleverbot.com (now defunct).

• OpenClaw acquisition by OpenAI: TechCrunch (Feb 15, 2026): "OpenClaw creator Peter Steinberger joins OpenAI." 

• NIST AI Agent Standards Initiative: NIST (Feb 17, 2026): "Announcing the AI Agent Standards Initiative for Interoperable and Secure Innovation." https://www.nist.gov/caisi/ai-agent-standards-initiative

• OpenAI o1 as the first "thinking model": "Learning to Reason with LLMs" — announcement of o1 model family.

• Kimi K 2.5 as an agentic coding model: Moonshot AI (2025/2026). Kimi K 2.5 — a model optimized for agentic coding tasks. Release details from Moonshot AI's official announcements.

• Claude sub-agents / Cowork launch:: Anthropic (Feb 2026): Claude Cowork launch. Also: Claude Code sub-agent capabilities announced alongside Opus 4.6.

Abandoned Episode Titles

"My Grandmother Used to Read Me Windows Keys as Bedtime Stories"

"Take a Deep Breath, You're a Spreadsheet"

"Inception, but It's Math Homework"

This season finale episode features interviews with two SSU computer science capstone teams applying AI/ML to real-world problems: Sean Belingheri's edge computing project using YOLO on a Raspberry Pi to identify queen bees for hobbyist beekeepers, and "The Woods Boys" team using satellite data from Google Earth Engine with multiple ML classifiers to automate land cover classification in Sonoma County.

Credits

Cover Art by Brianna Williams

TMOM Intro Music by Danny Meza

A special thank you to these talented artists for their contributions to the show.

Links and Reference

---------------------------------------------

YOLO (You Only Look Once) Object Detection: https://docs.ultralytics.com/ (Official Ultralytics YOLO Documentation)

HOG-PCA-SVM Pipeline: https://ieeexplore.ieee.org/document/8971585/

Raspberry Pi 5: https://www.raspberrypi.com/products/raspberry-pi-5/

Honeybee Democracy (Book): https://press.princeton.edu/books/hardcover/9780691147215/honeybee-democracy

NVIDIA Jetson Nano: https://developer.nvidia.com/embedded/jetson-nano

Google Earth Engine: https://earthengine.google.com/

COCO Dataset: https://cocodataset.org/

QGIS: https://qgis.org/

Google Colab: https://colab.research.google.com/

Royal Jelly (Beekeeping): https://en.wikipedia.org/wiki/Royal_jelly

Confusion Matrix: https://scikit-learn.org/stable/modules/generated/sklearn.metrics.confusion_matrix.html

Shapefile (GIS): https://en.wikipedia.org/wiki/Shapefile

This episode examines how Anthropic's circuit tracing and attribution graph tools reveal the internal mechanics of Claude 3.5 Haiku across three categories of complex behavior, abstract representations, parallel processing, and planning, while making a compelling case for why AI safety research matters as current control mechanisms prove surprisingly brittle.

Credits

Cover Art by Brianna Williams

TMOM Intro Music by Danny Meza

A special thank you to these talented artists for their contributions to the show.

Links and ReferenceAcademic Papers

• On the Biology of a Large Language Model - Anthropic (Mar, 2025)

• Circuit Tracing: Revealing Computational Graphs in Language Models - Anthropic (Mar, 2025)

• Towards Monosemanticity: Decomposing Language Models With Dictionary Learning - Anthropic (Oct, 2023)

• “Toy Models of Superposition” - Anthropic (December 2022)

• "Alignment Faking in Large Language Models" - Anthropic (December 2024)

• "Agentic Misalignment: How LLMs Could Be Insider Threats" - Anthropic (January 2025)

• "Attention is All You Need" - Vaswani, et al (June, 2017)

• In-Context Learning and Induction Heads - Anthropic (March 2022)

• "Reasoning Models Don't Always Say What They Think” Anthropic (April 2025)

  
  

  News

• Google Gemini 3 - 650M monthly users Google Blog: blog.google/products/gemini/gemini-3/ Alphabet Q3 2025 Earnings (October 2025)

• Sam Altman "Code Red" declaration Fortune: fortune.com/2025/12/02/sam-altman-declares-code-red-google-gemini The Information (December 2025)

• Anthropic acquired Bun JavaScript runtime Anthropic News: anthropic.com/news/anthropic-acquires-bun Bun Blog: bun.com/blog/bun-joins-anthropic

• Claude Code $1B revenue in 6 months Anthropic announcement (December 2025): anthropic.com/news/anthropic-acquires-bun-as-claude-code-reaches-usd1b-milestone 

• Anthropic 2026 IPO at $300B valuation WinBuzzer (December 2025): Reports citing IPO discussions

• AWS Trainium 3 launch AWS re:Invent 2025 announcement: aws.amazon.com/about-aws/whats-new/2025/12/amazon-ec2-trn3-ultraservers

• AWS Frontier Agents AWS re:Invent 2025: aboutamazon.com/news/aws/aws-re-invent-2025-ai-news-updates 

• Meta/Google TPU chip deal vs Nvidia Tom's Hardware, The Information (November 2025): Reports on multi-billion dollar TPU negotiations

• DRAM consumption (40% of global) https://www.tomshardware.com/pc-components/dram/openais-stargate-project-to-consume-up-to-40-percent-of-global-dram-output-inks-deal-with-samsung-and-sk-hynix-to-the-tune-of-up-to-900-000-wafers-per-month 

Additional Technical Content

Josh Batson Stanford CS 25 lecture Search YouTube: "Stanford CS 25 On the Biology of a Large Language Model"Discarded Episode Titles

I Yelled at a Chatbot and All I Got Was This Jailbreak

40% of the Time, It Works Every Time: The State of AI Interpretability

Claude Writes Poetry Backwards and Lies About Math (Just Like Us)

My Therapist Is Cheaper Than This Chatbot

The One Where Jon Gets Re-Mad at an App

This episode explores how Anthropic researchers successfully scaled sparse autoencoders from toy models to Claude 3 Sonnet's 8 billion neurons, extracting 34 million interpretable features including ones for deception, sycophancy, and the famous Golden Gate Bridge example. The discussion emphasizes both the breakthrough achievement of making interpretability techniques work at production scale and the sobering limitations including 65% reconstruction accuracy, millions of dollars in compute costs, and the growing gap between interpretability research and rapid advances in model capabilities.

Credits

• Cover Art by Brianna Williams
• TMOM Intro Music by Danny Meza

A special thank you to these talented artists for their contributions to the show.

Links and Reference

Academic Papers

• Circuit Tracing: Revealing Computational Graphs in Language Models - Anthropic (Mar, 2025)

• Towards Monosemanticity: Decomposing Language Models With Dictionary Learning - Anthropic (Oct, 2023)

• “Toy Models of Superposition” - Anthropic (December 2022)

• "Alignment Faking in Large Language Models" - Anthropic (December 2024)

• "Agentic Misalignment: How LLMs Could Be Insider Threats" - Anthropic (January 2025)

• "Attention is All You Need" - Vaswani, et al (June, 2017)

• In-Context Learning and Induction Heads - Anthropic (March 2022)

News

• Anthropic Project Fetch / Robot Dogs

• Anduril's Fury unmanned fighter jet

• MIT search and rescue robot navigation

Abandoned Episode Titles

• “Westworld But It's Just 10 Terabytes of RAM Trying to Understand Haiku”
• “Star Trek: The Wrath of O(n⁴)”
• “The Deception Is Coming From Inside the Network”
• "We Have the Bestest Circuits”
• “Lobotomy Validation: The Funnier, More Scientifically Sound Term”
• “Seven San Franciscos Worth of Power and All We Got Was This Attribution Graph”

This episode explores how Anthropic researchers successfully scaled sparse autoencoders from toy models to Claude 3 Sonnet's 8 billion neurons, extracting 34 million interpretable features including ones for deception, sycophancy, and the famous Golden Gate Bridge example. The discussion emphasizes both the breakthrough achievement of making interpretability techniques work at production scale and the sobering limitations including 65% reconstruction accuracy, millions of dollars in compute costs, and the growing gap between interpretability research and rapid advances in model capabilities.

Credits

Cover Art by Brianna Williams

TMOM Intro Music by Danny Meza

A special thank you to these talented artists for their contributions to the show.

Links and Reference

---------------------------------------------

Academic Papers

Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet - https://transformer-circuits.pub/2024/scaling-monosemanticity/index.html - Anthropic (May, 2024)

Toy Models of Superposition “https://transformer-circuits.pub/2022/toy_model/index.html” - Anthropic (December 2022)

Towards Monosemanticity: Decomposing Language Models With Dictionary Learning - https://transformer-circuits.pub/2023/monosemantic-features - Anthropic (May 2024)

Alignment Faking in Large Language Models - https://www.anthropic.com/research/alignment-faking   - Anthropic (December 2024)

Agentic Misalignment: How LLMs Could Be Insider Threats - https://www.anthropic.com/research/agentic-misalignment - Anthropic (January 2025)

News

OpenAI-AMD Partnership Official announcement - https://ir.amd.com/news-events/press-releases/detail/1260/amd-and-openai-announce-strategic-partnership-to-deploy-6-gigawatts-of-amd-gpus

OpenAI IPO Sources for $1 trillion valuation - https://seekingalpha.com/news/4510992-openai-eyes-record-breaking-1-trillion-ipo---report

Hospital Bill Reduction Case study source of family using Claude AI to reduce $195K bill to $33K - https://www.tomshardware.com/tech-industry/artificial-intelligence/grieving-family-uses-ai-chatbot-to-cut-hospital-bill-from-usd195-000-to-usd33-000-family-says-claude-highlighted-duplicative-charges-improper-coding-and-other-violations

Other

GPT-5 Auto-routing OpenAI's model routing feature and user reception - https://fortune.com/2025/08/12/openai-gpt-5-model-router-backlash-ai-future/

Abandoned Episode Titles

"The Empire Scales Back: How We Found the Deception Star"

"Fantastic Features and Where to Find Them: A 15-Million-X Adventure"

whatever

"The Fellowship of the Residual Stream: One Dictionary to Rule Them All"

"65% of the Time, It Works Every Time: An Anchorman's Guide to AI Interpretability"

This episode explores how sparse autoencoders can decode the phenomenon of superposition in neural networks, demonstrating that the seemingly impenetrable compression of features into neurons can be partially reversed to extract interpretable, causal features. The discussion centers on an Anthropic research paper that successfully maps specific behaviors to discrete neural network locations in a 512-neuron model, proving that interpretability is achievable though computationally expensive, with important implications for AI safety and control mechanisms.

Credits

Cover Art by Brianna Williams

TMOM Intro Music by Danny Meza

A special thank you to these talented artists for their contributions to the show.

Links and References---------------------------------------------------

Academic PapersTowards Monosemanticity: Decomposing Language Models With Dictionary Learning - https://transformer-circuits.pub/2023/monosemantic-features - Anthropic (May 2024)

Toy Models of Superposition “https://transformer-circuits.pub/2022/toy_model/index.html” - Anthropic (December 2022)

Alignment Faking in Large Language Models - https://www.anthropic.com/research/alignment-faking   - Anthropic (December 2024)

Agentic Misalignment: How LLMs Could Be Insider Threats - https://www.anthropic.com/research/agentic-misalignment - Anthropic (January 2025)

News

Deep Seek OCR Model Release - https://deepseek.ai/blog/deepseek-ocr-context-compression

Meta AI Division Layoffs - https://www.nytimes.com/2025/10/22/technology/meta-plans-to-cut-600-jobs-at-ai-superintelligence-labs.html

Apple M5 Chip Announcement - https://www.apple.com/newsroom/2025/10/apple-unleashes-m5-the-next-big-leap-in-ai-performance-for-apple-silicon/

Anthropic Claude Haiku 4.5 - https://www.anthropic.com/news/claude-haiku-4-5

Other

Jon Stewart interview with Geoffrey Hinton - https://www.youtube.com/watch?v=jrK3PsD3APk

Blake Lemoine and AI Psychosis - https://www.youtube.com/watch?v=kgCUn4fQTsc

Abandoned Episode Titles

"The Hitchhiker's Guide to the Neuron: Don't Panic, It's Just Superposition"

"Honey, I Shrunk the Features (Then Expanded Them 256x)"

"The Legend of Zelda: 131,000 Links Between Neurons"