📰 Not Boring → Nvidia underwrites the whole industry, the assistant moves into your health records, and the open-vs-closed border debate gets sharper

⌚️ Productivity Tapas → a PRD generator built from real customer calls, fully local Mac agents, and a nightly digest of what your team actually shipped

🍔 Blog Bites → why governance is the real product question, a concrete definition of product sense, and the case for taking a real sabbatical

🎙️ Pod Shots → #146, Own Your Intelligence Stack: eighteen conversations on who should own the AI your product runs on

OpenAI is close to landing a $500bn data centre with Nvidia's backing, and separately Nvidia is in talks to guarantee $250bn of financing for a second OpenAI data centre - on top of the $500bn deal, not instead of it.

Nvidia is also backing Ilya Sutskever's Safe Superintelligence with fresh compute commitments, extending the same playbook of buying influence over every serious frontier lab, not just OpenAI.

Data centres are on track to consume a fifth of all US power by 2035, and ASML shares slid on reports China is starting domestic DUV lithography tool production - the chip supply chain both labs depend on getting less exclusively Western.

Granola launched an Apple Watch app that transcribes meetings straight from your wrist, and X relaunched a fully rebuilt Android app after a year-long modernisation of scrolling, loading and notifications.

Meta quietly launched Pocket, an AI app that generates and shares interactive mini-games from text prompts - built on the team it acquired from vibe-coded gaming platform Gizmo - while Roblox's AI game-creation feature "Build" entered public alpha, letting anyone turn a prompt into a basic game with no code.

Whatnot acquired ML startup Shaped to power real-time recommendations for live shopping, and TechCrunch argues entertainment apps are converging into one universal format as AI makes it cheap to support every content type inside a single app.

Passionfroot raised $15m to expand its B2B creator marketplace into the US, and TechCrunch rounded up App Store "hidden gems" proving there's still room for genuinely good software in the AI era.

OpenAI made ChatGPT Health available to all US users and unveiled Presence, a platform for enterprises to launch realtime voice agents and chatbots - the same week it's also financing $750bn of data centre capacity to run all of it.

X Money officially launched in the US with Apple Wallet support after an invite-only beta, and Google now lets you log into your account with a selfie instead of a forgotten password.

Apple launched a product leasing programme through Klarna, and Meta gave Facebook Marketplace its own standalone app called Seller.

Waymo is reportedly mulling a breakup with Uber, a sign the robotaxi player thinks it no longer needs the distribution partner that got it to scale.

Silicon Valley is splitting over whether to close the borders to Chinese AI models, the same week Not Boring argues AI is oil, not god - a commodity input to be secured and priced, not a singular intelligence to be feared or worshipped.

An Anthropic model found flaws in tough-to-crack encryption algorithms, while Anthropic also published new rules for context engineering aimed at Claude 5-generation models - capability and the discipline for managing it, in the same week.

Travis Kalanick's robotics company raised $1.7bn led by a16z, and Tesla is training Optimus with employees at its Grünheide plant - both still solving the same "how do we get real-world data" problem 1X and Figure hit earlier this year.

India's first private rocket, Skyroot's Vikram-1, reached orbit - a genuine new-entrant milestone in a launch market that's mostly been a two-horse SpaceX/Rocket Lab race.

Why Software Factories Fail makes the case that most "AI-native" engineering orgs are optimising the wrong layer, while Cloudflare argues Durable Objects are made for agents - two takes on what infrastructure an agent-heavy stack actually needs underneath it.

The State of WebMCP surveys how fast the "agent-readable web" standard is actually spreading, and Sierra published a dev diary on building its MCP Gateway - production lessons from a company betting its whole product on agents talking to tools correctly.

Canvas UI shipped as an open-source component library built for AI-generated interfaces, and for the lower-level crowd, Everyone Should Know SIMD is a reminder that the performance fundamentals underneath all this agent tooling haven't gone anywhere.

BuildBetter: generates PRDs directly from real customer calls, support tickets and Slack threads, citing actual quotes and timestamps instead of generic AI prose.

Osaurus: open-source AI agents that run 100% locally on your Mac, no cloud dependency - for teams wary of sending sensitive product data to a third-party API.

Glaze by Raycast: build your own small Mac apps just by chatting with AI - no code required, useful for quick internal tools.

Eodly: reads Slack, GitHub and Linear each evening and sends one sourced report on what the team actually shipped versus what was claimed - handy for PMs managing multiple workstreams.

Governance protects mission from growth pressure: Costco's structural safeguards - staggered boards, poison pills - are the reason it stayed mission-led long enough to compound into a huge outperformance versus the S&P.

The Anthropic parallel is deliberate: Dario Amodei consulted Ries when founding Anthropic and implemented a Long-Term Benefit Trust for the same reason Price rebuilt Costco - to survive the stage where growth pressure corrupts mission.

Compromising compounds: Ries's line - "one small price increase hooks you into repeating it quarterly" - is the mechanism by which governance-light companies drift.

Tailor the structure, don't copy it: Public Benefit Corporation at seed, trust or mission-lock vehicle through Series A-C - the right vehicle depends on what you're protecting against.

Pattern-matching only works when you know its limits: Kahneman and Klein's three categories - high-validity, low-validity, and "wicked" situations where past patterns actively mislead - map directly onto product decisions.

Vicarious research isn't a substitute: studying products, reading reviews, and analysing competitors doesn't build product sense - only closing the full loop from decision to measured outcome does.

AI is a threat to this skill, not just a tool for it: if PMs stop owning decisions and outcomes end-to-end, they stop building the pattern library that product sense depends on.

Staying long enough to see results is the discipline: the loop only closes if you're still there when the outcome lands - a structural argument against constant team-hopping.

Two months is the floor, not the target: DiDonna's research suggests true disconnection takes longer than most people budget for; the real value shows up in months three through twelve.

Achievers risk burning out the break itself: chasing tangible accomplishments during a sabbatical can just relocate the burnout, not resolve it.

Boundaries have to be absolute: any consulting work or email-checking resets the clock on genuine disconnection.

The point is finding out if the path still fits: Lenny's own sabbatical led him to start this newsletter - the value wasn't the rest itself, it was the clarity that came from stepping away long enough to ask the question.

🏗️ Coding was Anthropic's identity crisis solved - Penn reveals the pivot from general-purpose chatbot to coding-first happened earlier and more deliberately than outsiders realise.

🔓 Open source AI converges as safety imperative - Horowitz, Jensen, and Kant independently conclude: monopoly AI is the real danger.

⚡ Building on frontier labs is getting dangerous - Horowitz warns Anthropic enters every successful app category and subsidises against its own ecosystem.

🎯 "Evals are the new PRDs" from two independent sources - Penn at Anthropic and Nitski at Mercor reach the same framework without citing each other.

🤖 Agent sprawl hit the wall - SaaStr went from 20+ agents to consolidation; the orchestration layer is where value moves next.

🧠 Context intelligence beats raw AI capability - Stripe's Kevin Yen and MATG's Grace both show the moat isn't the model - it's what you feed it.

📈 The ROI debate splits on timeline - Nitski sees no problem. Big Technology sees a ticking time bomb. IBM's 25% crash suggests the market is choosing sides.

🔧 Product craft endures but the toolkit shifts - Samsonov calls synthetic users a scam. Kevin Yen says MVPs can be more ambitious. Both are right.

🔮 Governance is becoming the product question - Ries frames Anthropic's velocity as governance-driven; Kant warns a two-company oligopoly is dystopian fiction made real.