💪 Mighty Brief
Thursday, March 5, 2026 — Generated 3:30 AM CT
⚡TL;DR — Top Stories
- Dario Amodei's leaked memo calls OpenAI Pentagon deal "safety theater," says DOD dislikes Anthropic for not giving "dictator-style praise to Trump" — The Information
- OpenAI hits $25B annualized revenue (up 17% from $21.4B at year-end), Anthropic narrowing gap — The Information
- OpenAI prepping IPO — hired Cooley + Wachtell, could come this year — The Information
- Jensen Huang: "OpenClaw is probably the single most important release of software" at Morgan Stanley TMT Conference — via @rowancheung
- Codex 5.3 (xhigh) solves 6-month bug in 45 minutes — Mitchell Hashimoto's report; Romain Huet amplifies — @mitchellh
- Alibaba shakeup: Qwen AI head resigns, new task force formed to accelerate foundation models — Techmeme/Reuters
- Simon Willison on Qwen 3.5 + Donald Knuth praises Claude Opus 4.6 for solving open problem — simonwillison.net
- Minnesota $3.7B budget surplus — from projected deficit to surplus — @BestPixMN
🏗️Blank Metal
📊 Pipeline Overview
| Stage | Deal | Status |
|---|
| Lead | UHG | HOT Training — Anthropic ref |
| Lead | Oshi Health | HOT Anthropic ref, $500K-$1.5M |
| Lead | Tungsten Auto | HOT Anthropic chain, discovery next wk |
| Lead | SRI | WARM Intake — qualify |
| Discovery | NeoGov | HOT $150K-$1.5M |
| Discovery | Talkiatry | HOT Proposal Fri, $400M+ company |
| Discovery | BBG | HOT Stalled — samples 19d late, $391K |
| Discovery | Bento | HOT Blocked — Jeff ends Fri, access 19d late |
| Discovery | Bond Sports | HOT Scoping scheduled |
| Approach | PointsNorth | HOT Proposal this wk, feasibility done |
| Approach | Insurance Newco | HOT Demo review, March call |
| Approach | Caribou | HOT ~$40K sprint, MSA in review |
| Proposal | PaceLoanGroup | HOT ~$100K, 3 wks no reply |
| Proposal | Dasvandh | WARM $125-170K, 23 days waiting |
| SOW | PowerPusher | WARM $78-135K, SF blocker 3.5wk |
| SOW | Zivian Health | WARM Partner POV needed |
| SOW | Pivotal | WARM Reviving, $36-50K |
| — | Valvoline | COLD 36d stale |
| — | hc1/SourceIQ | COLD 35d silent — archive? |
| — | Marqi Medical | COLD Since Oct 2025 |
👥 Team Activity (24h)
- Git: Granola auto-sync only (3 commits) — no client-facing code pushes
- Teresa: Emailed re: participant guide and introductions (Turn River Cowork prep)
- Teresa: Google Slides comment tagging Dan on Cowork Training deck
- Tufin: meetings folder updated (Julien dry run prep)
⚡ Dan's Plate Today
- 9:00 AM — PowerPusher ↔ Blank Metal regroup (w/ Elli, rblesi)
- 10:00 AM – 12:00 PM — Busy block
- 12:00 PM — The Smelter (engineering knowledge sharing) + Weekly Headlines
- 12:15 PM — Scoping & Proposal Approach: Part 1 — Dan presents Dasvandh approach
- 1:00 PM — Mark & Dan checkin (overlaps with Turn River dry run)
- 1:00 PM — Turn River Cowork Training Dry Run (Teresa + Jackie producing)
- 2:30 PM — BlankMetal :: TurnRiver sync (Zoom w/ eginsburg@turnriver.com)
- 3:30 PM — Tufin / Julien Legendre — Dry run of Claude Cowork setup screen sharing
📧 Teresa email needs reply: participant guide & introductions for Turn River. No other urgent BM emails.
🔗Capture Themes (last 48h)
Agent-Per-Client Is the New SaaS
Captures: @hnshah (SaaS morphs to agent-per-client), @karrisaarinen (Linear parallel agents), @danshipper (OpenClaw guide — Every team adopted), @DanHightowerJr (agent startup launching real businesses as demo)
Why it matters: The SaaS→agents transition is happening in production, not theory. BM's model (deploy specialized agents per client) is exactly this pattern — and OpenClaw is being cited as the deployment mechanism.
Healthcare AI Window Is Wide Open
Captures: a16z/Julie Yoo × Ambience ($30M net new margin, 75%+ clinician adoption), AI&I/Boulton & Watt (Moxie — business-in-a-box for nurse med spas, 600+ customers), @jasonlk (dinosaur founders need AI agents)
Why it matters: Healthcare AI is generating real margin, not just pilots. Ambience's numbers validate the BM healthcare thesis. Moxie IS the Lite Medical model — exact same play for nurse-owned med spas.
The Agent Setup IS the Moat
Captures: @snowmaker (Claude Code = rocket for new repos), @systematicls (world-class agentic engineering), 0x01f (AI-era moats), @Alfred_Lin (legendary company playbook), @nikunj (per-user model divergence as lock-in)
Why it matters: Compound engineering is converging as consensus. The moat isn't the model — it's the accumulated agent infrastructure, behavioral data, and team workflow. This is T3 being validated from five independent sources.
Ship Now or Be Disrupted
Captures: @gregisenberg (most asymmetric moment in history), @jasonlk (dinosaur age — pull your best 10-20 people), Claude Catharsis (The Information — life's work becomes free), Bill Gurley/Founders #413 (run down a dream)
Why it matters: The urgency thesis (T1/T3) keeps getting louder. Lemkin's prescription — isolate top talent, sole mission = build the #1 AI agent in your vertical — is literally BM's pitch to clients. The window is now.
🧠Munger Observer — March 4
Opportunity Cost — The Kitchen Tour: Full morning on Cowork training metaphors. High-value for BM positioning, but that's time NOT on Pulltab geocoding fix, Squorum, or sketch-to-code. Ship the metaphor and move on before the perfection tax compounds.
Second-Order — Pulltab's Trust Window: Jay called with two bugs. Speed of fix = signal about whether Pulltab Magic is a real product. The $5 Google geocoding fix has asymmetric payoff.
Circle of Competence — Today's Calendar: 7+ meetings across ventures. Which are inside your circle (building, architecture) vs. outside (proposal formatting, dry runs)? The Smelter and Tech PowWow are core. Delegate more aggressively.
"Invert, always invert."
📧Newsletters & Email
- The Information: OpenAI Tops $25B Annualized Revenue — read
- The Information: OpenAI Selects Law Firms for IPO Prep — read
- The Information: OpenAI Scales Back Shopping Plans for ChatGPT — read
- The Information: OpenAI Held Early Talks With Trade Desk to Sell Ads — read
- The Information: Read Anthropic CEO's Memo Attacking OpenAI — read
- The Information: A Claude Catharsis: When Your Life's Work Becomes Free — read
- The Pragmatic Engineer: Building Claude Code with Boris Cherny — read
- Lenny's Newsletter: Three Job Searches, Three AI Roles: What Actually Worked — read
- Every: How Claws Took Over Every — read
- Sahil Bloom: 25 Powerful Paradoxes of Life
🐦X / Twitter
AI List
- Jensen Huang on OpenClaw: "Probably the single most important release of software" — @rowancheung
- Codex 5.3 xhigh solves 6-month bug in 45 min for $4.14 — @mitchellh / @romainhuet
- Amjad Masad: AI compressing how we build — roles collapse, roadmaps expire quickly — @amasad
- Replit team grinding late nights — 9pm "early" — @amasad RT @nickco
- Peter Dedene: GPT-5.4 hinted? "May the fourth... 5.4 👀" — @dedene
- roon: Military AI future — rapid OODA loops, swarms vs large assets, offense favored — @tszzl
- WaPo details on military Claude usage — @tszzl RT @kevinschaul
- NASA Force announced under Tech Force — @NathanLands
Leaders List
- Nikita Bier: "This startup idea is fundable now" (image) — @nikitabier
- Garry Tan: Coding agents + 🦞 catalyzing open access to closed systems via CLI — RT @illscience
- Garry Tan: Garry's List launch + citizen's union — @garrytan
- Hiten Shah: LightSail shipped something timely and good — @hnshah
Tech-Dev List
- Deveillance Spectre I — anti-surveillance device, "productized tin foil hat" — @mgsiegler
- Ruby best language for Claude Code (benchmark by @mametter) — types hinder AI, token efficiency matters more — @paolino
- Grammarly "Expert Review" simulates feedback from famous writers without permission — Techmeme/Wired
- Alibaba Qwen task force after AI head resignation — Techmeme/Reuters
- Unleash raises $35M Series B — AI risk management in dev — Techmeme
- Perplexity Computer running entire GTM — @AravSrinivas RT
Twin Cities List
- Minnesota $3.7B surplus — from projected deficit, expect complaints either way — @BestPixMN
- MN fraud: Officials aware of fraud concerns since 2019, billions stolen, warnings ignored — @andybrehm / NRO
- $75M Anoka mental health facility — 50-bed inpatient expansion — @BusinessMN
- Flow Environmental $20M factory — "mechanical room in a box" — @TCBmag
- Bombs strike Tehran — heaviest since war began — @marrina RT Telegraph
Tools List
- Cloudflare: Dynamic User Risk Scores in Access policies — behavioral detection (impossible travel, etc.) — @eastdakota
- Cloudflare sandbox hint: Matthew Prince teasing something re: sandbox spin-up speed — @eastdakota
- Peter Steinberger: Also amplifying Jensen/OpenClaw quote — @steipete
- Interaction: Auto-archiving tool — @interaction
📡RSS / Blogs
- Simon Willison (Mar 4): "Something is afoot in the land of Qwen" — Qwen 3.5 coverage + team departures (Alibaba shakeup) — read
- Simon Willison (Mar 3): Donald Knuth on Claude Opus 4.6 solving open problem + Gemini 3.1 Flash-Lite (1/8th price of Pro) — read
📰Headlines
Techmeme
- Lead: Dario Amodei memo — "safety theater" — dominates top of page
- Alibaba new task force for foundation AI after Qwen head departure
- Grammarly AI "Expert Review" simulates famous writers
- Capcom: PC now 50%+ of total unit sales, expects increase
- Unleash $35M Series B — AI risk management in dev
- Munich's UnternehmerTUM tops FT startup hub ranking 3rd year
Memeorandum
- Lead: GOP Sen. Steve Daines (Montana) won't seek re-election — surprise last-minute withdrawal
- Heavy strikes on Tehran — hospitals, schools, residential
🧩Trending Skills (24h)
- agent-tools (tul-sh/skills) — #1 trending, 14.8K installs
- twitter-automation (tul-sh/skills) — 14.8K
- qwen-image-2 (tul-sh/skills) — Qwen image gen, 14.8K
- remotion-render (tul-sh/skills) — video rendering, 14.7K
- ai-image-generation (tul-sh/skills) — 14.6K
- nano-banana / nano-banana-2 — Gemini image gen variants
- agent-browser — browser automation for agents, 14.6K
- ai-video-generation — 14.4K
- find-skills (vercel-labs) — meta-skill for discovery, 14.0K
Notable: Microsoft's github-copilot-for-azure skills (azure-ai, storage, observability, deploy) all trending at ~9.6K — enterprise AI infra tooling wave.
💭Personal Thought
Jensen calling OpenClaw "the single most important release of software" while Dario leaks a 1,600-word rage memo about Pentagon politics — that's the split screen of AI in March 2026. One side is infrastructure becoming essential. The other is geopolitics eating the labs. Your captures this week keep circling the same point from different angles: the agent setup is the moat, the window is closing, ship now. You've got 7 meetings today across BM, and yesterday's Munger Observer flagged exactly this: circle of competence. The Smelter and the Turn River prep are core. The rest? Delegate harder. The "Working in Public" open loop is still the biggest gap — you've been collecting signal for months. At some point the capture has to become the output.
Generated by Mighty · mighty-brief.pages.dev · Archive