Do Not Disturb My Circles AlphaFold cost under $1M to train. OpenAI spends $2.3B on inference. The chatbot era consumed the talent and compute that could have cured diseases.

Worth reading: Google DeepMind's CEO on the future he didn't get, a CERN for AI that could have led to cures, instead of a commercial arms race for chatbots.

philippdubach.com/posts/do-not-disturb-my-...

The Geometry of Who Knows What When neither side can define the states of the world, adverse selection fears are misplaced. Zeckhauser

Worth reading: investing at the edge of knowledge, specifically how to gauge if the other side of a trade has an information edge, and what happens when that assumption breaks down.

philippdubach.com/posts/the-geometry-of-wh...

Why Lilly's Weight Loss Pill Isn't a Peptide Oral semaglutide destroys 99% of its active ingredient per dose. Lilly

Worth reading: Eli Lilly skipped peptide tech for its weight loss pill, bypassing a $1.8 billion problem Novo Nordisk solved, but with limited results.

philippdubach.com/posts/why-lillys-weight-...

Ambiguity by Design Ellsberg proved people flee unknown odds. Zeckhauser showed their flight creates mispricing. Part 2 on ambiguity aversion, comparative ignorance, and investing.

Investors flee from uncertainty, not just risk. An experiment where people consistently made irrational choices when probabilities were unknown versus known has lasting implications for financial markets.

http://philippdubach.com/posts/ambiguity-by-design/

Three Kinds of Not-Knowing Knightian uncertainty splits not-knowing into risk, uncertainty, and ignorance. A century after Knight and Keynes, most of investing still ignores the split.

David Ricardo made a fortune buying British bonds before Waterloo not because he was informed, but because he understood his own ignorance. A useful distinction in investing is that not all not-knowing is created equal.

philippdubach.com/posts/three-kinds-of-not...

The iPhone 17's on-device AI model is impressive but consumers don't care about specs. Parameters could be the new megahertz or megapixels - a marketing focus that might just work.

philippdubach.com/posts/on-device-ai-model...

AI Can Now Design Drugs in Seconds; We Still Can't Tell You If They Work. IsoDDE doubles AlphaFold 3 on hard benchmarks and beats physics-based gold standards. But no AI drug has FDA approval. What $4B in pharma deals actually mean.

Worth reading: AI can design drugs in seconds but pharma partnerships reveal a 50:1 ratio between promised milestone payments and actual upfront costs, suggesting a cautious bet on unproven tech.

philippdubach.com/posts/ai-can-now-design-...

The Last Architecture Designed by Hand The transformer

Sam Altman says the transformer isn't the final form, and he's right. The math isn't in its favor. A new architecture could be on the horizon.

philippdubach.com/posts/the-last-architect...

MCP vs A2A in 2026: How the AI Protocol War Ends MCP leads with 97M monthly SDK downloads and 10,000+ servers. A2A fills a different layer. Analysis of the agentic AI standards war with historical parallels.

Sam Altman endorsed a rival's standard, MCP, on March 26, 2025, which may be the most significant event in enterprise AI infrastructure this year, indicating the protocol war is close to over.

philippdubach.com/posts/mcp-vs-a2a-in-2026...

AI Models Are the New Rebar Qwen 3.5-35B runs on a gaming PC and matches Claude Sonnet 4.5. When the commodity version is 95% as good and 97% cheaper, you have a pricing problem.

Alibaba's Qwen 3.5 model runs on a $2,000 GPU, matching top-tier performance at a fraction of the cost, with a 97 percent discount on input tokens compared to Claude Sonnet 4.5.

philippdubach.com/posts/ai-models-are-the-...

AI Capex Arms Race: Who Blinks First? Alphabet

Big tech's AI spending spree is projected to consume 94% of operating cash flow minus dividends and buybacks. At what revenue growth rate does this pay for itself?

philippdubach.com/posts/ai-capex-arms-race...

93% of Developers Use AI Coding Tools. Productivity Hasn't Moved. METR found experienced developers 19% slower with AI, despite feeling 20% faster. At 92.6% adoption, organizational productivity gains remain roughly 10%.

Worth reading: a study of 16 experienced developers found AI coding tools made them 19% slower, despite their own perception of being 20% faster.

philippdubach.com/posts/93-of-developers-u...

Peter Thiel's Physics Department Peter Thiel says physics stalled in 1972. Then GPT-5.2 proved a new result in theoretical physics. The 75:1 AI compute gap between commerce and science.

Worth reading: Peter Thiel's stagnation thesis, that progress in the physical world has stalled since the 1970s, got a boost from an unlikely source - AI proving a long-standing problem in theoretical physics.

philippdubach.com/posts/peter-thiels-physi...

When AI Labs Become Defense Contractors The Anthropic-Pentagon standoff isn

Worth reading: How quickly AI labs can become defense contractors, and the financial gravity that shifts their priorities.

philippdubach.com/posts/when-ai-labs-becom...

When Every Bulge Bracket Bank Agrees on AI I read 12 AI research reports from Goldman Sachs, JPMorgan, UBS, and 6 other banks. Here

Worth reading: big banks agree AI is a general-purpose tech like electrification, but quietly diverge on forecasts, with estimates ranging from +0.7% to +10-15% impact on productivity.

philippdubach.com/posts/when-every-bulge-b...

People Live in Levels, Not Rates Prices rose 25% since 2020 and won

Jon Stewart's interview with Richard Thaler highlighted a disconnect between economics and societal well-being, with Stewart surprisingly stumbling into a standard economic solution while trying to reject it.

philippdubach.com/posts/people-live-in-lev...

The Absolute Insider Mess of Prediction Markets A Google insider made $1.15M on Polymarket in 24 hours. Israeli soldiers bet classified strike timing. Why prediction markets need insider trading regulation.

Someone close to Google made $1.15 million in under 24 hours on Polymarket, a prediction market, by betting on 23 Google-related outcomes and getting 22 right. The regulatory response so far has been zero.

philippdubach.com/posts/the-absolute-insid...

Novo Was Europe's Most Valuable Company Novo Nordisk lost 75% since June 2024. CagriSema failed vs Zepbound, US pricing is resetting lower, and Lilly leads on every axis. Full breakdown with numbers.

Novo Nordisk lost 75% of its value since June 2024, falling from Europe's most valuable company to behind 10 others, after a key drug trial failed and US pricing pressures mounted.

philippdubach.com/posts/novo-was-europes-m...

Super Bowl Ad Economics 2026: The $8M Prisoner's Dilemma A 30-second Super Bowl spot costs $8M. The real price is $16–23M. The ROI evidence is mixed. A deep look at the pricing, the prisoner's dilemma, and the NFL.

The economics of a Super Bowl ad, costs and viewership.

philippdubach.com/posts/the-ec...

Thanks a lot! Really appreciate the feedback!

The Impossible Backhand AI converges to the mean by design. On Humanity

AI generated content hits a ceiling due to convergence to the mean, a structural limitation.

philippdubach.com/posts/the-impossible-bac...

Europe's $24 Trillion Payment Breakup Is Really a Bet on Infrastructure Arbitrage The EuroPA alliance connected 130 million users across 13 countries overnight. But this isn

Europe's payment initiative is a bet on infrastructure arbitrage, not just a breakup with Visa and Mastercard.

philippdubach.com/posts/europes-24-trillio...

Your Manager Is Not the Problem 71% of executives have witnessed managers sabotaging talented employees. Harvard research shows status threat, not money, is the primary driver, and subjective evaluation discretion nearly doubles sabotage rates. The problem is incentive design, not character.

New research suggests badly designed systems, not bad managers, cause sabotage in the workplace.

philippdubach.com/posts/your-manager-is-no...

The Long Volatility Premium One River

The Long Volatility Premium challenges conventional wisdom on buying puts as a losing game.

philippdubach.com/posts/the-long-volatilit...

The SaaSpocalypse Paradox The market is simultaneously pricing AI capex failure and AI destroying all software. Both cannot be true. A data-driven analysis of the February 2026 enterprise software sell-off, the hyperscaler sustainability question, and why Goldman Sachs, a16z, and the All-In pod see a 20x TAM expansion where the market sees extinction.

The SaaSpocalypse paradox, where markets simultaneously price AI failure and destruction of software, cannot hold true.

philippdubach.com/posts/the-saaspocalypse-...

Don't Go Monolithic; The Agent Stack Is Stratifying The enterprise AI agent stack is stratifying into six layers with different winners at each. Models commoditize; context — your organizational world model — compounds. A framework for agentic AI architecture decisions.

Enterprise AI agent stack is stratifying into layers with different winners at each level.

philippdubach.com/posts/dont-go-monolithic...

Where Mobile Money Goes Now Apps beat games in mobile revenue for the first time in 2025. GenAI led the charge with $3.5B in growth. Eight charts show who controls the app economy.

Non-game mobile apps now generate more revenue than games, led by AI and subscriptions.

philippdubach.com/posts/where-mobile-money...

The Variance Tax Variance drain is the hidden cost of volatility: why a portfolio averaging +10% can lose money. The ½σ² formula explains the gap between paper and real returns.

The variance tax reduces investment returns due to volatility, calculated as half the variance of returns.

philippdubach.com/posts/the-va...

Claude Opus 4.6: Anthropic's New Flagship AI Model for Agentic Coding Claude Opus 4.6 brings a 1M token context window, 68.8% ARC-AGI-2, and Agent Teams to Claude Code. Full benchmark comparison vs GPT-5.2 and Gemini 3 Pro with pricing analysis.

Anthropic releases Claude Opus 4.6, its new flagship AI model for agentic coding.

philippdubach.com/posts/claude-opus-4.6-an...

Buying the Haystack Might Not Work This Year a16z sees AI fundamentals thriving with 80% GPU utilization. AQR sees the CAPE at the 96th percentile. Both have data. Both may be right.

Andreessen Horowitz and AQR reports on the market offer conflicting views.

philippdubach.com/posts/buying-the-haystac...