💪 Mighty Brief

Friday, February 27, 2026 — Travel Day (Aruba → Minneapolis)

📌 TL;DR — Top Stories

  1. Google strikes multibillion-dollar TPU deal with Meta — Meta renting Google's AI chips to train models, a direct shot at Nvidia's dominance. Meta also scrapped its most advanced internal training chip. The Information
  2. Block cuts ~4,000 employees (nearly half workforce) — Jack Dorsey targeting $2M+ gross profit per person, 4x pre-COVID efficiency. Stock up 24%+ after hours. Jack's response
  3. Netflix walks away from Warner Bros. — Paramount wins — WBD board deemed Paramount's $31/share bid superior. NFLX up 8%. David Ellison's mega-merger taking shape. CNBC / Techmeme
  4. Bezos' Project Prometheus raising tens of billions — Buying companies hit by AI disruption. Valued ~$30B. Techmeme
  5. DOD vs. Anthropic heats up — Pentagon in talks with AI cos to automate reconnaissance of China's power grids. Anthropic holding firm on no autonomous weapons / no mass surveillance. Could lose $200M contract by Friday. FT / Ben's Bites
  6. Karpathy: programming changed more in 2 months than years — Step function since December. Lemkin: "If your startup hasn't gotten wildly more productive since December, something is wrong." Lemkin
  7. Jared Friedman: dev tools need API-first account creation — "In the Claude Code age, Claude can't sign up on its own." Agent-first API design is now tablestakes. Link
  8. NFX: "What comes after SaaS?" — Deploying intelligence at scale. Not bigger SaaS — completely different game, 100x stakes. NFX Newsletter

🏗️ Blank Metal

📊 Pipeline

Recent git activity (last 24h): Cowork training sessions planned (Open Book Mar 4, Tufin Mar 6, Tern River Mar 8), Dan/Elli 1:1 notes filed, PE portco AI ops playbook added, Highmark CakePop AI inquiry framework created, Open Book research updated.

DealStatusNotes
Open Book🟢 ActiveTraining Wed Mar 4, curriculum revised, connector research done
Tufin🟢 ActiveTraining Fri Mar 6 (virtual), Dan facilitating
Tern River🟢 ActiveTraining Mon Mar 8 (virtual), Dan facilitating
Highmark🟢 ActiveCakePop AI inquiry framework (Scott/Jason)

👥 Team Activity (last 24h)

⚡ Dan's Plate

🧠 Munger Observer

Opportunity Cost: Vacation day with 10+ workstreams touched. The breadth is impressive but tomorrow's a travel day — what needed depth today that got breadth instead? The Liz Cumby brainstorm and Keith geospatial meeting are both high-leverage relationships that reward focused follow-through over context-switching.

Second-Order (Narrative Gap): Koylan's Brain OS thread got 2.2M views describing architecture Dan already built. "Ahead on architecture, behind on narrative" — every week of delay means someone else's framing becomes the default. The "File System Is the Character Sheet" post isn't content marketing, it's competitive positioning.

Incentive Analysis: Keith at Lifetime is "surprisingly far along" on AI dashboards. Orgs that build internal AI first are the hardest to sell to and the best to partner with. The Wick Embed frame (partner, not vendor) matches the incentive structure correctly.

"Invert, always invert."

🔗 Capture Themes (last 48h)

1. Agent-First Product Design

Captures: @petergyang (build for agents first), @danshipper (personal software = agents you have relationship with), @gregisenberg (AI-native vertical SaaS playbook), Simon Willison (agentic engineering patterns)

Why it matters: This is the _M training thesis made real — Cowork plugins, agent-first APIs, and "the user in UX is becoming an agent" all validate the consulting play Dan's running next week.

2. The December Inflection

Captures: @trengriffin/Karpathy (programming step-function change), @rohanpaul/Uber CEO (90% of engineers on AI, top 30% unprecedented productivity), @levie (weekend agents as normal), @paulg (techno-turbulence good for startups)

Why it matters: Validates compound engineering thesis (T3) and the window narrative. Every enterprise client _M trains is proving this — the ones who lean in now compound, the rest fall behind.

3. The Anthropic/Pentagon Tension

Captures: Ben's Bites (Claude conflicts — responsible scaling policy revised, DOD access denied), @lulumeservey (Pentagon losing info war to Anthropic), @tszzl/roon (open chaos as democratic strength)

Why it matters: Anthropic's position on military use is forcing a public reckoning on AI governance. If you're training enterprises on Claude, this is context they'll ask about.

📧 Newsletters

🐦 Twitter — AI

🐦 Twitter — Leaders

🐦 Twitter — Tech/Dev

🐦 Twitter — Tools

🐦 Twitter — Twin Cities

📰 Headlines

Techmeme

Memeorandum

Featured Podcasts

🧩 Trending Skills

Top trending on skills.sh (24h): agent-tools, remotion-render, agent-browser, ai-image-generation, nano-banana, ai-video-generation, twitter-automation. Microsoft's Azure AI/Observability Copilot skills climbing fast.

Agent-browser and agent-tools trending #1-3 — the "agents use computers like humans" wave Ben's Bites flagged is showing up in skill installs too.

💭 Personal Thought

You're flying home from vacation into three back-to-back Cowork trainings (Open Book, Tufin, Tern River). Meanwhile, Karpathy says programming changed more in 2 months than in years, Friedman says dev tools can't even let agents sign up for accounts yet, and Block just proved you can 4x efficiency per head with AI.

The Munger Observer nailed it yesterday: you're ahead on architecture, behind on narrative. These three trainings are live proof of the thesis — but they're also consulting revenue, not content. The "File System Is the Character Sheet" post keeps not getting written while Koylan's version gets 2.2M views. What if one of those travel-day hours on the plane became the draft?

💪 Mighty Brief — Generated at 3:30 AM CT

← Archive