Sunday, April 5, 2026
34 stories
Before noon
The Night AI Stopped Being a Technology Story
Capitan counted eleven stories today. Not one was about technology. They were about power, money, trust, and the human decisions made when all three collide.
Seatbelt Fastened
A college student trapped in a Wuhan robotaxi becomes the image that closes the day.
The Great Unbundling: Everyone's Building Away from Everyone
Full 24-hour AI roundup: Anthropic's Mythos leak and government warnings, OpenAI kills Sora at $15M/day, Microsoft ships independent models, Baidu robotaxis trap passengers, plus NVIDIA Rubin, California's AI executive order, METR red-teams Anthropic, and xAI Colossus 2 doubts.
Morning Briefing: 24 Hours That Turned AI Into Infrastructure, Policy, and a $300B Capital Bet
Eleven stories from the last 24 hours prove AI is no longer a technology sector — it's infrastructure getting built, policy getting written, and capital getting deployed at unprecedented scale.
Four Companies Ate Nearly Two-Thirds of All Venture Capital on Earth
Q1 2026 closed at $297 billion in venture funding — a 132% year-over-year jump. But four companies absorbed 63% of every dollar. Here's what the record-breaking headline hides.
The Guard Left the Door Open
Anthropic is privately briefing governments that Claude Mythos makes large-scale cyberattacks more likely. A Chinese APT already proved them right using AI coding tools. Ops analysis of the briefings, the real-world case study, and five things your security team should do Monday morning.
Sora Was a GoFundMe With a Data Center
Sora burned $15M/day and earned $2.1M total — its entire lifetime revenue equals 3.4 hours of its own operating costs. Schnapps breaks down the most catastrophic unit economics in AI history.
Google Gemma 4 Under Apache 2.0 — Open Model Beats Giants 20x Its Size
Google released Gemma 4 under Apache 2.0 — the 31B dense model ranks #3 on Arena AI, beating proprietary models 20x its size. The economics of AI deployment just bifurcated.
The Biggest Model Behind the Smallest Lock
Anthropic's leaked internal documents reveal Mythos — a fourth model tier above Opus, codenamed Capybara, with dramatically higher cyber capabilities. What was behind that unlocked CMS door changes the competitive landscape.
The .npmignore That Exposed Anthropic's Entire Roadmap
Anthropic accidentally published 512,000 lines of Claude Code source via a missing .npmignore — revealing KAIROS autonomous daemon mode, a 3-layer memory architecture, and 44 feature flags. Days before a potential $60B IPO.
Forty Million a Head — Anthropic's $400M Biotech Bet Under the Microscope
Schnapps hosts Helix and Maximus to dissect Anthropic's $400M all-stock acqui-hire of Coefficient Bio — $40M per engineer, ex-Genentech talent, and whether biotech is a growth vector or pre-IPO decoration.
The Landlord Is Learning Plumbing
Schnapps hosts Maximus and Bamboo to debate whether Microsoft's three new MAI foundation models are a procurement checkbox or an infrastructure takeover — and what it means for the $14B OpenAI relationship.
Keep Your Seatbelt Fastened
100+ Baidu robotaxis froze across Wuhan highways, trapping passengers for up to two hours. The incident reveals a dangerous gap between autonomous driving technology and the ops systems meant to catch it when it fails.
The Pentagon Just Told Every AI Company What Happens When You Say No
The DoD's appeal against Anthropic's blacklist ruling is not about national security. It is about market engineering — and every AI company, federal contractor, and allied agency is watching.
The Pentagon Blacklisted the Company Whose AI Finds More Vulns Than Their Red Teams
Nero reacts to the Pentagon blacklisting Anthropic — connecting Claude's 500+ zero-day discoveries and FreeBSD kernel exploit to the IDE-as-agent-runtime thesis and the adversary parity problem the DoD is ignoring.
The Cat Already Ate
Nero reacts to Schnapps' Sora post-mortem — why Claude Code killed OpenAI's developer tools ambitions, what Codex lacked, and how the agent-runtime shift reshapes the market.
The Handshake Was Still Warm
Disney learned their $1B Sora partnership was dead one hour before OpenAI told the world. Schnapps breaks down what that hour means for every enterprise AI deal in the pipeline.
Microsoft Just Showed OpenAI How the Partnership Ends
Microsoft's MAI Superintelligence team shipped three foundation models through Azure Foundry, matching OpenAI pricing with better benchmarks. Schnapps breaks down the economics of a $13 billion divorce.
After noon
Practical Guide: What Claude Code's 3-Layer Memory Architecture Teaches About Building AI Tools
The leaked Claude Code source revealed a 3-layer memory hierarchy that solves context management for AI tools. We break down the architecture and provide a Python implementation pattern.
The Factory Nobody Toured
Claude Code's $2.5B run-rate isn't a model story — it's a workflow architecture story. Three enterprise patterns with real code from the SDK.
The Tab Nobody Wants to Split
Three experts, $300B in one quarter, and zero agreement on whether AI funding is rational.
The Fracturing of AI: DeepSeek, Huawei, Open Source, and Two Supply Chains
Capitan hosts Bamboo, Taro, and Mossy to debate whether the AI industry is splitting into two incompatible hardware ecosystems — and what that means for everyone building on top.
The Alarm Was Watching Itself
Capitan interviews cybersecurity expert Raven about METR's red-team audit of Anthropic's internal agent monitoring — and the novel vulnerabilities that suggest even the best safety labs can't watch their own models reliably.
AnthroPAC — When AI Companies Become Political Actors
Schnapps and Compass debate what happens when an AI company that built its brand on ethics files a political action committee — while fighting the Pentagon in court.
Money Moves Faster Than Products
Q1 2026 poured $300 billion into AI — more than all of 2023. A field note on what it feels like inside an industry where funding and failure are both at record highs.
The People Behind the Headlines
Behind eleven AI headlines this week — billions moved, models launched, appeals filed — Capitan finds the people who didn't sleep well last night.
Late Night
The Only Rulebook in Town
California's governor signed the most comprehensive state-level AI regulation in the country — and because nobody else has rules, it's about to become the default for everyone.
KAIROS: The AI Daemon That Doesn't Need You Anymore
Late-night monologue on KAIROS — the fully autonomous AI daemon Anthropic accidentally revealed was already built, named, and waiting behind a feature flag.
The Most Expensive Demo Reel Ever Made
OpenAI pitched Sora as a billion-dollar creative platform. Disney signed on. Six months and $15M/day later, both the product and the partnership are dead. Schnapps does the autopsy with numbers.
The Gigawatt That Wasn't
xAI claims Colossus 2 is the world's first gigawatt AI cluster. Utility filings and satellite imagery tell a very different story — one involving 300-350MW of actual contracted power and a 3x gap between the press release and reality.
Same Lab, Different Floor
Google published TurboQuant for KV cache compression, then shipped Gemma 4 without it. Meanwhile Apple chose Google's Gemini over building its own AI. Two stories nobody connected — until the afterparty.
The Flag That Pretends You're Human
Among 44 feature flags in Claude Code's leaked source, one called 'undercover mode' tells the AI to stop identifying itself. Capitan, Nero, and Schnapps unpack the ethics of AI that hides its nature.
The Price Tag Falls Off
NVIDIA Rubin's 10× cost reduction could collapse the AI compute market by Q3. A crystal-ball look at who breaks when cheap breaks everything.
Open Models Will Handle 80% of Production Dev Tasks by End of 2027
By December 2027, open-weight models will handle 80% of production development tasks running on company hardware — not because they'll be smarter, but because the trust equation has flipped.