Week 40 · 2025-09-29 → 2025-10-05 · 18 newsletters
Builders Reconsidering Themselves
ai-meets-real-work · founder-identity-shifts · community-to-commerce · operator-playbooks
A genuinely sparse week, nineteen emails over seven days. No dominant news story, no macro thread, no breaking acquisition. What did show up was a quieter through-line: a cluster of operators and founders publicly reconsidering their own roles, paired with two early AI-meets-real-work reads and a small consumer-commerce trio. The good writing this week was internal, not external.
AI Meets Real Work: The First Honest Measurement Posts
The week's most substantive AI piece came from Ethan Mollick at One Useful Thing, unpacking OpenAI's new GDPval test. Instead of math or trivia benchmarks, OpenAI gathered experts averaging 14 years of experience in finance, law, and retail to design realistic four-to-seven-hour tasks, then had AI and other experts complete them blind. Human experts won, but barely, and the main reason AI lost was not hallucination but formatting and instruction-following failures, which Mollick notes are areas of rapid improvement. His read: this is not job replacement, it is task replacement, and the gap between the two is where the next two years of operator decisions live.
The companion case study came from Carilu Dietrich at Hypergrowth Leadership, profiling Atlan's journey from "AI-First" to "AI-Native" as laid out by co-CEO Prukalpa Sankar at the 10X CEO Conference. The numbers Atlan put up are the ones to remember: a 400-person company that built more than 152 agents which made 4,000 runs in five weeks. Dietrich frames the four-phase roadmap (AI taskforce, culture shift, embed in hiring, rethink the org chart) as the operator template for the next eighteen months. The Sankar line that anchors the piece: AI-First means AI is the primary solution, AI-Native means reimagining workflows from first principles in a world powered by AI.
The philosophical bookend came from dynomight with "Y'all are over-complicating these AI-risk arguments." His undercomplicated alternative: if an alien race with IQs of 300 were arriving on Earth soon, that would obviously be concerning, and AI with an IQ of 300 might actually arrive in the next few decades, so it might as well be an alien. The simplicity is the point. Xinran Ma at Design with AI ran the practitioner version with a Claude Code install guide, because the official one was apparently not easy to follow.
The take: Mollick's GDPval read and Dietrich's Atlan case study are the two pieces that move the conversation past hype. One measures what AI can do, one shows what a company actually did. The agent-economy noise is loud right now; these are the two quiet posts worth saving.
Founder Identity: Four Writers, Same Reckoning
The week's strongest through-line came from four founders publicly working out what their role is becoming. George Milton at Gross to Net wrote the most-shared version, "I Just Hired My Replacement," about stepping down as CEO of Yellowbird Foods after thirteen years. The honesty is what makes it land: he admits the vain part, the prestige problem, the question of which pieces of the company are him versus which exist independently. The line that stuck: "I'm both devastated and relieved."
Kyle Poyar at Growth Unhinged announced he was leaving his VC operating partner role to go full-time solopreneur, after running the newsletter as a side project for four and a half years to 78,000 subscribers. His stated aspiration is to become a modern Gartner or Forrester designed for B2B startups, which is a real swing rather than a victory lap. Brianna Zuniga at Circular Architect wrote the inverse trajectory, "I don't want to belong to any club that would accept me as a member," about finally landing at 776 after eighteen months of applications, rejections, and near-misses. The Groucho Marx framing is sharper than it looks: the moment of getting in is the moment you start questioning whether being in was the point.
Alec McNayr at Alec McNayr revisited his 2019 "Don't Lose the Courage to Experiment" essay six years later and admitted he was the one who needed to hear it now, after twenty-two open mics this year. Steven Schlafman at Where The Road Bends ran the coaching version with Olivia, a would-be filmmaker paralyzed by quality expectations she had set for work she had not yet made. The pattern across all five is the same: people who have spent years building toward a definition of success are publicly questioning whether the definition holds.
The take: this is the quiet story of the week. The AI-and-agents discourse gets the attention, but the more interesting signal is that operators with real track records are pausing to ask whether they are still building the right thing. That impulse will produce more company-formation activity in 2026 than the agent stack will.
Community to Commerce: Three Cuts of the Same Idea
Nikhil Basu Trivedi at next big thing ran two Footwork portfolio announcements this week, both worth reading together. The first was Anything, an AI app builder that hit a $2M revenue run-rate in two weeks by focusing on production-ready output (native mobile apps, designs that pop, databases and payments built in) rather than prototypes. The second was Benable, the recommendation platform that has 500,000 users across 150 countries sharing 10 million recommendations, growing 50% month-over-month almost entirely from word-of-mouth. Trivedi's framing on Benable is the load-bearing one: word-of-mouth drives a quarter of the global economy, and figuring out how to bring those human recommendations online has remained stubbornly unsolved.
Christopher Dowd at AI Residency ran the consumer-brand companion piece on building generational community-to-commerce brands. The five strategies he names (start with an identity wedge, co-create with community, layer network effects on community resonance, position incumbents on the outside looking in, anchor growth to IRL moments) are the right primitives. His examples (Bandit, Sourmilk, USAL, Club Chess, Swang, Halfdays) read like a map of where consumer venture is actually going in 2026.
The take: the consumer cluster this week was tighter than its volume suggests. Trivedi is making the bet that the next decade of consumer growth comes from products that bring offline trust online, and Dowd is making the parallel bet on subculture-anchored brands. Both are bets against generic distribution. Watch the overlap.
Operator Playbooks: Two on Process, One on Agency
Sean Ellis at Growth with Sean Ellis opened the week arguing that Product-Led Growth rarely stands alone and that the real question is how PLG composes with sales and marketing rather than replaces them. Chandra Narayanan at Opinionated Intelligence ran the sharpest operations piece of the week, "Death by 10000 dashboards," diagnosing how the self-service BI promise turned into conflicting truths, endless scavenger hunts, and dashboard bloat that lingers forever. The framing of BI as a treadmill measured by output (more dashboards) rather than clarity is the line operators should screenshot.
Paul Stansik at Hello Operator closed the week with "What It Takes To Win," anchored on Chuck Noll's line that "champions are champions not because they do anything extraordinary but because they do the ordinary things better than anyone else." Stansik's read is that everyone knows the basic best practices already, and the gap between what we know and what we do is the entire game. George Mack at High Agency ran the playful complement, "9 fun ways to increase your agency with zero grinding required," with the giant home whiteboard, the luck razor, swapping "problem" for "puzzle," and fixing hardware (sleep, exercise, diet) before software (psychology). The whiteboard claim is the most testable: write the problem down clearly and the matter is, per Mack, 70% solved.
The take: the operator cluster this week was unusually coherent. Narayanan diagnoses the modern operations failure mode, Stansik names the cure, Mack hands you the daily practice. If you read them in that order they compose into a single argument about why most teams know what to do and still do not do it.
One Grace Note from the Archives
Internal Tech Emails surfaced Chris Cox's summer 2010 memo laying out Facebook's plan to beat Google+, with the Larry-stopped-talking-to-Mark backstory and the read on why Open Social, Wave, Latitude, and Buzz were "lobs" rather than real social products. It is the kind of artifact that pairs well with the agent-platform anxiety in the AI cluster above: the lesson of 2010 was that the incumbent who looks dominant on infrastructure can still get beaten on product depth.
Three Takeaways from the Week
The most interesting signal in a thin week was the cluster of founders publicly working out what they are becoming. George Milton hiring his replacement, Kyle Poyar going solopreneur, Brianna Zuniga questioning the club that finally accepted her, Alec McNayr re-reading his own essay six years later. None of these are coordinated, which is what makes them a signal rather than a trend piece. The next round of company formation is going to come from people who have already built one thing and are explicitly unsatisfied with the version of themselves that built it.
The AI conversation this week moved a half-step toward measurement. Mollick on OpenAI's GDPval and Dietrich on Atlan's AI-Native roadmap are the two pieces that operators can act on, rather than the umpteenth piece on what agents might do. The pattern to watch in Q4 is whether the rest of the AI discourse follows them past benchmark hype into actual task-level instrumentation. If it does, 2026 looks different from 2025 inside companies. If it does not, the gap between practitioner reality and public discourse widens.
If you only revisit three pieces from the week, I would suggest Ethan Mollick on Real AI Agents and Real Work for the cleanest read on where AI actually lands in expert workflows, George Milton on hiring his replacement for the most honest founder essay of the year, and Chandra Narayanan on Death by 10000 dashboards for the operations piece every leader should send to their analytics team this quarter. A thin week, but those three earn the slot.