๐ช Mighty Brief
Sunday, March 8, 2026 โ 3:30 AM CT
โก TL;DR โ Top Stories
- Iran war escalates โ Strikes near Tehran airport, Iran apologizes to Gulf neighbors while missiles still fly. Trump expanding targets. Kashkari warns economic uncertainty complicates Fed. NYT Live
- Pentagon vs Anthropic heats up โ Emil Michael profiled by Bloomberg as the face of the dispute. Anthropic's 20-page military restrictions, Palantir incident, designated supply chain risk. Techmeme
- Codex 5.4 reverse engineers DOS game โ Ammaar Reshi ran Codex for 6 hours to decompile SkyRoads with zero source code, rebuild in Rust. Open-sourced all prompts. Thread ยท Repo
- OpenClaw 2026.3.7 โ GPT-5.4 + Gemini 3.1 Flash-Lite support, pluggable context engines, ACP bindings survive restarts. 196 contributors. Release
- Atlassian CEO on SaaS Apocalypse โ 3-bucket framework: seats tied to replaced work (โ zero), seats as headcount proxy (fine), middle ground. Vibe coding overstated, extensibility is the real play. a16z Pod
- Sam Lessin vibe-codes AI podcast twin โ molchat.ai: credible AI-generated More or Less episodes in <30 min of partial attention. molchat.ai
- GPT-5.4 Pro physics leap โ Greg Brockman shares research-level physics reasoning improvements. Tweet
- Agent observability raises $44M โ GV-led seed + Series A at $300M valuation for AI agent dev/deploy/observe platform. Techmeme
๐ Capture Themes (last 48h)
AI in the Kill Chain
All-In E263 (Emil Michael) + Bloomberg profile + Stratechery CSIS interview
Anthropic's military policy is the inflection point for AI governance. Pentagon designated them a supply chain risk after an exec called Palantir about classified ops. All models asymptoting, but policy and trust differentiation is the new moat. Direct relevance to _M: healthcare AI ethical boundaries face similar scrutiny.
Operators Building Their Own AI
Sam Lessin (molchat.ai) + Ammaar Reshi (Codex SkyRoads) + Garry Tan ("coding like 2013")
T4 validated repeatedly: the operator who understands their domain deeply builds faster than any agency or dev shop. Lessin spins up a credible AI podcast twin in 30 min. Ammaar reverse engineers a DOS game from binary. Build velocity is absurd and accelerating.
SaaS Repricing Under AI
Atlassian CEO 3-bucket framework + Levie on cheaper software + agent observability $300M raise
T1/T2 evidence in real-time. Cannon-Brookes names the categories plainly: some SaaS goes to zero (support tools), some gets stickier (HR/payroll), the middle is a bet. Meanwhile the picks-and-shovels layer (agent observability) is getting bid up fast.
๐๏ธ Blank Metal
๐ Pipeline: No git activity in the last 24h (weekend). Leads directory has ~25 prospects tracked. Recent activity in Bond-Sports (Mar 3), Dasvandh-Network (Mar 2), Oshi-Health (Feb 25), SRI (Feb 25).
๐ฅ Team: No commits in last 24h. Last notable: dry-run meeting notes from Mar 6.
โก Dan's plate: No unread _M emails. No _M meetings today (Sunday). Dry run prep notes from Thursday โ review for any follow-up items Monday.
๐ Recent meetings: Mar 6 dry run โ appears to be presentation rehearsal. Check for action items when back in the office.
๐ง Munger Observer
Opportunity Cost โ 9 of 14 drafts cooling. Only 2 drafts have momentum, 9 are 14+ days stale. The chains feeding them are loaded โ the-question-changed has 6 fresh captures this week alone. The raw material compounds; the drafts don't write themselves. Every week of delay is decay, not patience.
Incentive Check โ Training others on your tools. A 4-hour Claude Cowork for Turn/River is high-touch delivery. Does training a client to use the tool you built create a moat, or does it commoditize your edge? If the value is in the orchestration layer (your thesis), then teaching execution-level tool use may be the wrong altitude.
Inversion โ What breaks if you ignore the 4 broken crons? Gmail triage, podcast monitor, markdown health, skill directory โ all errored today. None catastrophic alone. But compounding failures in automation infrastructure is exactly how you wake up one morning with no morning briefing and no idea why. Fix or kill.
"Invert, always invert."
๐ง Newsletters
- Exponential View #564 โ "Intelligence as a target; the future of knowledge; AI, productivity & economy; CO2 armor." Opens with "The Strait of Compute" โ critical infrastructure strikes in Bahrain context. Read
- Tanner Tomassi Pickleball โ Christian Alshon teaches forehand attack (PPA #3). Gmail
- Yashar Ali โ "On Jew Hatred and Trolls." Gmail
๐ฆ Twitter โ AI List
- @emollick โ AI fiction writing is a reader-interpretation problem. AI generates "meaning-like" metaphors. "The return of the humanities!" Thread
- @ammaar โ Codex 5.4 reverse-engineered SkyRoads (DOS) to Rust. 6 hours, no source code. Open-sourced prompts + progress. Demo
- @tszzl (roon) โ "Claude is tasteful in a high reddit modernist way, GPT is tasteful in an early twitter schizophrenic way." New GPT is "some sort of postrationalist." Tweet
- @amasad โ RT: Anthropic's Claude helped select targets for Iran opening strikes. Tweet
๐ฆ Twitter โ Leaders
- @garrytan โ "i'm coding like it's 2013 again." Also RT'd Levie on software getting vastly cheaper to write. Tweet
- @eladgil โ Girard quote on desire: "If it came from our deepest selves, it would not be desire." Tweet
- @hnshah โ RT'd OpenClaw 2026.3.7 and "Datadog for humans" โ real-time heart rate correlated with typing speed/app usage. Tweet
- @garrytan โ Housing thread: hard left becoming slightly more YIMBY but math doesn't add up. Tweet
๐ฆ Twitter โ Tech/Dev
- @Techmeme โ Emil Michael profile (Bloomberg) re Pentagon/Anthropic dispute. Link
- @Techmeme โ Samsung "open to strategic co-operation" with more AI groups, adding Perplexity. Link
- @Techmeme โ Agent dev/deploy/observe platform raises $44M at $300M (GV-led). Link
- @jakemintz โ "When I don't know what I want: Claude. When I want something correct: Codex." Tweet
- @fkpxls (Tina He) โ Netbird: share local projects with one command, secure public URL. Tweet
๐ฆ Twitter โ Tools
- @browser_use โ Launched /loop: persistent agent loops, not limited to 3 days. Tweet
- @gdb (Greg Brockman) โ GPT-5.4 Pro (xhigh) making big jumps in research-level physics reasoning. Tweet
- @jxnlco โ Demo of Codex driving Ableton for music production. Tweet
- @openclaw โ 2026.3.7: GPT-5.4, Gemini 3.1 Flash-Lite, pluggable context engines, HEIF support. 196 contributors. Release
- @steipete โ The Verge attended ClawCon NYC. Tweet
๐ฆ Twitter โ Twin Cities
- State Hockey โ Moorhead beats Minnetonka 5-4 in 2OT to repeat as state champs. "Absolute bananas" โ@nacholeber. Tweet
- @BusinessMN โ Kashkari says Iran conflict raises economic uncertainty, complicates Fed rate decisions. Link
- @BusinessMN โ Federal panel delays vote on Trump's $400M White House ballroom after tens of thousands of opposing comments.
- First Avenue โ Peter McPoland + Girl Tones last night; Ella Red sold out in the Entry.
๐ฐ Headlines
Techmeme
- Big Technology Podcast: AI Revenue Explodes, Dario's Memo, McDonald's CEO's Baby Burger Bite
- Hard Fork: OpenAI's Fog of War + Betting on Iran + Review of Slop
- Access: What happens to Google when AI answers everything? (Liz Reid)
- Stratechery: Anthropic and the Military (CSIS interview) + Higher Powers and Lower Macs
Memeorandum
- Strikes land near airport in Iranian capital โ NYT/Reuters live updates
- Iran's president apologizes for strikes on neighbors even as missiles target their cities โ AP
- Trump says Iran targets will expand, honors troops killed โ multiple outlets
Drudge
- Inside War-a-Lago, makeshift Palm Beach situation room โ Vanity Fair
- Stranded Americans say they were abandoned โ WSJ
- 'Operation Epstein Distraction': Trump's bloody hype videos โ Guardian
- What life is really like under fire in Dubai โ Telegraph
๐งฉ Trending Skills
Top movers on skills.sh (24h):
- agent-tools (toolshell) โ 13.7K installs, #1 trending
- agent-browser (toolshell) โ 13.5K, browser automation skill
- nano-banana-2 (toolshell) โ 13.4K, image gen
- remotion-render (toolshell) โ 13.3K, video rendering
- frontend-design (anthropics) โ 2.9K
- vercel-react-best-practices (vercel-labs) โ 2.9K
- shadcn (shadcn/ui) โ 2.3K
Relevant to Dan's stack: agent-browser and agent-tools track with Mighty's automation layer. remotion-render interesting for content generation.
๐ RSS / Blogs
- Simon Willison (Mar 7) โ "Codex for Open Source" โ OpenAI matches Anthropic's offer: 6 months free ChatGPT Pro + Codex for OSS maintainers. Read
- Stratechery (Mar 7) โ Weekly: Anthropic and the Military (CSIS interview), Higher Powers and Lower Macs (Apple M5 bet). Read