Week 29 · 2025-07-14 → 2025-07-20 · 14 newsletters

Sparse Inbox, Loud Throughlines

ai-design-and-distribution · creative-careers-and-craft · human-in-the-haze

Mid-July, 17 emails across seven days. A quiet week with two clusters that actually held together (AI as a design and distribution problem; creative-career artifacts and craft) and a thread of personal writing from people sitting in their own haze. No macro news story dominated. The signal was operator-level and personal-level, in roughly equal measure.

AI as a Design and Distribution Problem, Not a Demo

The week's most coherent throughline ran across four pieces that all refused to treat AI as a feature to bolt on. Sophie Bakalar at AI Residency opened the week with A screenless future, a Gin Lane lunch and a line from Emmett Shine that did most of the work: "Humans existed without screens for hundreds of thousands of years. They will exist without screens for hundreds of thousands more." She mapped the bets being made against that future: Meta's AR/VR and Ray-Ban and neural wristband spend, OpenAI's $6.5B Jony Ive device, the new wave of stealth hardware startups. Three days later her colleague Christopher Dowd ran The evolving distribution game in AI, a debrief of 225+ AIR Cohort One applications. His read: the best founders treat distribution as part of the product, prototype go-to-market with feature-level discipline, and use Discord and Reddit as home base before they touch a paid channel. David Protein at the bodega checkout and Icon narrating its own growth on Instagram were the example brands.

Max Mitcham ran Prompt engineering Vs Context engineering, which made the operator-level case that prompt engineering has become a subset of the larger discipline of architecting the entire information ecosystem around an AI interaction. His framing: in production, the prompt matters less than the context, because the agent needs prior conversation, customer history, support tickets, and the rest of the surrounding world to do the job. Carly Ayres at Good Graf ran the design-side version in Should designers train models?, a profile of Justin Ouellette (Muxtape, Vimeo, Tumblr Special Projects, NYT R&D Lab, now Output) treating AI as a design material with real constraints rather than a shiny feature. The Russ Maschmeyer tweet ("Designers should learn to train models") was the prompt; Ouellette's response was the more interesting half: skeptical, curious, hands-on, irritated mostly at thoughtless deployments.

A counterweight came from Sayash Kapoor at AI as Normal Technology in Could AI slow science?, the week's most provocative single piece. His argument: even if individual scientists benefit from AI adoption, the system effects could be net negative because science is not a market and its production-progress paradox is already severe. Paper output grew 500x between 1900 and 2015 while actual progress, by most measures, stayed constant or slowed. AI accelerating the production side without fixing the progress side is the failure mode he flags.

The take: the AI conversation this week sat one rung above the hype cycle. Distribution as product (Dowd), context as architecture (Mitcham), AI as design material (Ayres), and AI as a systemic shock to science (Kapoor) are four different operator frames, and they share a posture. None of them treat the model as the interesting object. They treat the surrounding system as the interesting object. That is the right move, and it is the move most of the AI press still has not made.

Creative Careers, Job Artifacts, and the Mechanics of Getting Picked

A second cluster, almost accidental, formed around the question of how creative people get and keep work. Ben Kassoy at A Strawberry Spinning Like a Dreidel ran I'm seeking a (full-time!) job, a direct ask to his subscribers for VP of Copy, Head of Copy, Creative Director (Copy), or editorial leadership roles in LA, NY, or remote. Former DoSomething EIC, nine coauthored books, an award-winning solo show, the receipts are there. Three days later Carly Ayres ran the systemic version with her husband Sebastian in The artifacts you need to get hired, pulling from the professional practice course they taught at Parsons and SVA. Sebastian's framework: a website (your digital calling card), a presentation (your narrative for the room), a PDF portfolio (for follow-up and sharing), plus a resume wedged in as a necessary evil. Each artifact serves a different audience and moment, all working toward the same goal of getting someone to want a conversation with you.

Craig Zingerline ran Importance of retention in startup growth, the operator-level reminder that early-stage teams obsess over acquisition while later-stage teams obsess over retention, and that strong retention is what makes growth actually sustainable. TM at The General Partnership launched The General Podcast, a hostless format pairing peers in honest conversation rather than running yet another partner-interviews-founder show. Episode one paired Katie Dill (Stripe Head of Design) with Randy Hunt (Notion Head of Design) on building taste into products and what designers uniquely bring in the age of AI.

The take: the through-line is that creative and product work in 2025 is increasingly judged on artifacts and systems rather than on talent claims. Kassoy's open ask works because he has a portfolio and a track record; Ayres's framework works because students who built the three artifacts landed at Robinhood and Code and Theory and elsewhere; Zingerline's retention argument works because the artifact of a retained user is more credible than the claim of a great product. The market is asking for evidence, not pitch.

The Human in the Haze: Three Writers Doing It Out Loud

The third cluster was the most surprising, because it was the most consistent. Three writers, three different domains, sitting in personal turbulence and writing through it without hiding the turbulence.

Brianna Zuniga at Circular Architect ran i almost forgot that this is the whole point, a post about how writing her weekly "What the Water Carries" essay had become a spiritual anchor in a stretch of personal, professional, and family turbulence after the Camp Mystic floods and a Seattle trip. Her central image (each of us a first-timer at being human, mostly pretending otherwise) is the kind of line that earns the slot. Piera Luisa Gelardi at NoomaLooma ran Color Collecting Walks To Shift Your Mood, a creative-play practice piece about assigning a color to a feeling and going out into the city to collect it. JOLT-OF-ENERGY ORANGE: a Vespa, tangerine flowers, a construction barricade, an oversized roadside orange. "Soft fascination" is the psychology term she leans on, and the practice is unironically useful.

The two posts from signull sat in the same emotional register from the founder side. why we share: the architecture of sharability and the opposite of love isn't hate. it's "meh." both ran in the same week, the second documenting an early demo to two respected people whose involuntary, real smiles confirmed something the founder needed to feel. The lower-case voice, the deliberate slowness, the willingness to publish the small win, all of it is its own kind of operator practice.

Justin Mares at The Next ran Ray Peat and his high-sugar diet, but the more interesting frame in the piece was not the Peat material. It was his argument that most value comes from compounding, told through a friend running a four-person firm wondering whether to keep a fine-but-not-loved hire. Mares's read: anything that increases the risk you do not love running the company for 20+ years is a risk that kills the compound. The decision sounds small; the math is not.

The take: the personal-essay writers this week were doing the same thing the operator writers were doing, just in a different register. Both were arguing that the surrounding system (emotional state, founder energy, the long arc of compounding) matters more than the discrete artifact. Zuniga's anchor, Gelardi's color walk, signull's small smiles, and Mares's compounding argument are the same instinct: protect the upstream input, and the output takes care of itself.

One More Worth Noting

The Last Bear Standing ran The Ugly Duckling, a long pitch on Bridger Aerospace (BAER), the Belgrade, Montana wildfire-fighting aerial fleet founded by former Navy SEAL Tim Sheehy (now US Senator from Montana). The setup: 97% revenue CAGR over seven years, freshly cash-flow positive, going-concern warning dropped, and new legislation and budgets in the past eight weeks set up to accelerate the industry. Capital-intensive, dangerous, ugly, and tipping the right way. The fact that the founder is a sitting senator is the kind of detail that either matters a lot or not at all, and the piece is honest that the answer depends on how you read the next two years of fire-season budgets.

Two pieces did not slot cleanly into any cluster but were each their own thing: Zach Everson at 1100 Pennsylvania ran Trump's crypto firm raised $52 million, reporting on World Liberty Financial's second token sale distributing $50.7M to Trump and other founders rather than retaining proceeds inside the company, plus the Justin Sun memecoin commitment. JJ Chou ran a US cost-of-living analysis (in Mandarin) breaking down a Beaverton, Oregon family's expenses against a Taipei baseline.

Three Takeaways from the Week

The strongest writers this week refused the demo frame. Bakalar on screenless hardware, Dowd on distribution-as-product, Mitcham on context as architecture, Ayres on AI as design material, and Kapoor on AI as a systemic shock to science were all doing the same thing: treating AI as a system question rather than a feature question. That posture is rarer than it should be, and it is the posture that ages well.

The personal writing this week was doing the same work the operator writing was doing. Protect the upstream input. For Zuniga that was a weekly writing practice; for Gelardi a color walk; for signull a willingness to publish small wins out loud; for Mares the math of compounding over 20+ years. The discrete artifact is downstream of the surrounding system, and the writers who landed best were the ones who said that out loud.

If you only revisit three pieces from the week, I would suggest Sayash Kapoor's "Could AI slow science?" for the cleanest contrarian frame on AI's systemic effects, Carly Ayres on "The artifacts you need to get hired" for the cleanest operator-level take on creative-career mechanics, and Brianna Zuniga's "i almost forgot that this is the whole point" for the post most worth reading slowly in a week that mostly let you. A 17-email week is not a small week if you pick the right three.