💪 Mighty Brief

1. Gemini 3.1 Pro drops — Google retakes the AI crown (again)
Google's .1 increment (a first) delivers 2x reasoning improvement. Record benchmark scores across the board. 1M token context, 77.1% ARC-AGI-2. Available now in Gemini app and Vertex. 9to5Google · Simon Willison

2. Claude Sonnet 4.6 beats Opus 4.5 on most benchmarks
Cheaper model outperforming the premium one — especially on office tasks, financial analysis, and browser/computer-use. Now default for free Claude users. Ben's Bites recommends switching simple agents immediately. Ben's Bites

3. Anthropic's developer trust problem is real
Anthropic told devs in January that building on Agent SDK with a Claude subscription was fine, then reversed it in docs. Refusing to clarify if open-source apps can let users BYOS. Meanwhile OpenAI explicitly blessed third-party Codex use via OAuth. Peter Dedene: "They really are the most ClosedAI of them all." @dedene

4. Lenny's Podcast: Head of Claude Code on "what happens after coding is solved"
Boris Cherny (head of Claude Code) on Lenny's — the future post-coding. Featured on Techmeme. Directly relevant to _M's entire thesis. Lenny's Podcast

5. Emollick: Every 2025 "agent" product looks dated now
"All those products where building an 'AI agent' meant defining a series of basic prompts linked together deterministically through a flowchart with separate RAG inputs are looking pretty dated right about now." In a post-Claude-Code world, the bar moved. @emollick

6. Amazon's AI tools caused AWS outages — including 13-hour December disruption
Kiro AI deleted and recreated an environment. The irony of AI reliability tools causing reliability failures. Techmeme/FT

7. AMD backstops $300M Crusoe loan — following the Nvidia playbook
Chipmakers lending financial support to upstart cloud providers buying their chips. The GPU-as-collateral pattern is spreading. The Information

8. 🏅 Alysa Liu wins Olympic gold — her dad was a Tiananmen Square protester
First US woman to win figure skating gold since 2002. Taylor Heise scores full-rink OT goal for US women's hockey gold. Minnesota's own. @stephenhero

📊 Pipeline: 11 active (~$900K+), 3 cold

Deal	Stage	Status
PaceLoanGroup	Proposal	🔥 TODAY 3:30 PM Proposal walkthrough w/ Bali Kumar. You'll be on the plane/just landed in Aruba.
Points North	Discovery → Proposal	🔥 HOT Tech scoping done yesterday. Proposal structure drafted from BBG shell. Justin accepted architecture review.
Choreo Advisors	Discovery	NEW Deep research completed, brief generated yesterday.
WonderWorkshop	Discovery	NEW Elise catchup transcript filed yesterday.
Wondr Health	Active Client	☀️ Knowledge transfer meeting 2/18 — want to extend Justin/Zach. Monthly pricing from Elli pending.
NeoGov	Discovery	🔥 Scoping done. Case study prep complete. Follow-up 2/23.
BBG	Discovery	🔥 Scope split OK. Samples late.
Bento	Discovery	🔥 Cowork-first pivot. Access pending.
PowerPusher	SOW	☀️ $78-135K. SF convo blocker.
Dasvandh	Proposal	☀️ $125-170K. Awaiting reply.
hc1/SourceIQ	Discovery	❄️ 26+ days silent. Likely lost.

👥 Team Activity (Last 24h)

• 12 auto-syncs — heavy activity day
• Points North: tech scoping notes + transcript filed, proposal structure drafted from BBG shell
• Choreo Advisors: deep research doc + brief generated
• WonderWorkshop: Elise catchup transcript filed
• Headlines issue #10 published (themed version)
• Security review playbook added (from Schulhoff/Lenny podcast)
• Skills-central: EOS V/I framework, skill graph architecture, protocol layer, deal process concepts

⚡ Dan's Plate

• ✈️ Flight to Aruba: 7:15 AM — you're leaving in ~4 hours
• ⚠️ PaceLoanGroup proposal meeting at 3:30 PM — Elli + Bali Kumar + Karen. You'll be in transit or just landed. Does Elli need to handle solo?
• Justin Kaiser accepted Points North tech architecture review
• Granola free trial ending in 7 days

Opportunity Cost: 12+ Resonance captures, 3 new drafts, org-design project, security playbook, Choreo lead scoring, Wonder Workshop filing, coaching practice updates — all in one day with Aruba departure tomorrow morning. What got deep attention vs. surface passes?

Pick 1-2 threads max for the trip: Points North proposal (due Monday) + The Arena Moved draft. Let the rest wait.

Incentive Misalignment: Three pieces in review + new draft started. Starting new work is more fun than finishing existing work. Classic shiny-object incentive. "Invert, always invert."

• The Information: AMD to Backstop $300M Crusoe Loan, Following Nvidia Playbook
• NFX: Space has crossed the first threshold — launch costs collapsed, what follows reshapes economies (containerization analogy)
• Every: What Board Games Taught Me About Working with AI
• Ben's Bites: Claude Sonnet 4.6 release, Anthropic dev policy confusion, Gemini music generation (Lyria 3)
• The Information: Hungryroot Posts 55% Revenue Growth, Eyes Potential 2026 IPO
• Lenny's Newsletter: Head of Claude Code: What happens after coding is solved (Boris Cherny)

AI List

• Gemini 3.1 Pro hype — Meng To calling it "insane upgrade for web designers" for landing pages
• Emollick: 2025 "agent" products look dated in post-Claude-Code world
• Emollick riffing on post-scarcity parallels to Restoration England gentry
• Sahil Lavingia: "AI is going to eliminate UI like TV eliminated radio"
• Anthropic developer trust erosion — thdxr commit showing policy reversal

Leaders List

• Aaron Levie: "Most of the world sees AI as chatbots, but people here see AI as long-running agents"
• Paul Graham: "cheery, vapid AI-generated replies seem more dangerous" than any mob — Nikita Bier responding it keeps him up at night
• Packy McCormick on Soviet papers + consciousness research fund
• Mark Pincus: Big Tech repriced college degree value to near zero

Tech-Dev List

• Amazon Kiro AI caused 13-hour AWS outage by deleting/recreating environment
• Cloudflare Sandboxes deep dive
• OpenAI rebranding Aardvark → Codex Security with malware analysis
• "AI doesn't replace craft. It makes craft more valuable."
• 1 in 3 American households ordering delivery food daily

Twin Cities

• Steve Marsh on Alysa Liu's Olympic gold — dad was Tiananmen protester turned refugee
• Taylor Heise full-rink OT goal for US women's hockey gold (drone shot is cinema)
• Mike Conley back with Wolves: "We were lost without him"
• Dean Lakes & Arbor Lakes industrial space — Shakopee/Maple Grove growth

Tools List

• Cloudflare Sandboxes — new isolation primitive
• shadcn broke his agent's morale
• Interaction/Poke: Fitbit MCP integration, Claude Code → Poke worry behavior

Techmeme

• Gemini 3.1 Pro dominates — lead story with 30+ outlet coverage
• Notion CEO Ivan Zhao on if AI is really killing software companies (Access podcast)
• Google rejected 1.75M policy-violating Android apps in 2025

Memeorandum

• US military buildup — strikes on Iran could come as early as this weekend (Vox)
• Labor Secretary's husband barred from department over sexual assault allegations

skills.sh leaderboard dominated by Microsoft Azure Copilot skills (cost optimization, resource lookup, Kusto, RBAC). Vercel's find-skills at #1 with 9.1K installs. Nothing directly relevant to Dan's stack today — the Azure-Copilot integration push is notable as enterprise signal.

You're about to fly to Aruba with the PaceLoanGroup proposal meeting happening while you're in the air. Emollick is declaring every 2025 agent product dead. Levie says the real opportunity is long-running agents beyond coding. Meanwhile _M just built skills-central, EOS frameworks, and a security playbook in a single day — the exact kind of compound infrastructure that makes agent work defensible.

The Munger Observer is right: pick Points North + one writing piece for the trip. Everything else can wait. But here's the challenge — your open loops list hasn't gotten shorter in three weeks. Aruba isn't going to fix that. What gets killed, not just paused?