Week 37 · 2025-09-08 → 2025-09-14 · 22 newsletters

Operators Versus Optimists

ai-as-normal-tech · management-and-growth · thinking-traps

A quiet week. Twenty-five emails across seven days, with no breaking news cluster and no dominant industry thread. What did show up, though, sorted cleanly into three buckets: a maturing argument about how to frame AI's actual trajectory, a set of operator-level posts on management and growth that all circled the same point about leaner teams, and a small batch of writers asking what counts as clear thinking in the first place. The signal was thin but consistent.

AI as Normal Technology: The Frame Sharpens

The week's most substantive piece was Arvind Narayanan's relaunch of his newsletter as AI as Normal Technology, a rebrand away from the AI Snake Oil framing. The essay is a clarification of the original thesis: "normal" does not mean mundane or predictable, it means that AI's societal effects will play out the way other powerful technologies have, with unpredictable emergent consequences and a diffusion timeline measured in years rather than months. The contrast he sets up against AI 2027 is the most useful frame in the piece. If GPT-5's reception nudged you toward the normal-technology camp, Narayanan's argument is that you probably do not yet understand the thesis well enough.

The companion piece came from Ethan Mollick at One Useful Thing, "On Working with Wizards," which makes a softer but related point. Mollick argues that the co-intelligence framing from his book is starting to give way to something different: we are moving from partners to audience, from collaboration to conjuring. He fed his book and 140 posts into NotebookLM, asked it to make a video about what has happened in AI since publication, and the result was good enough that he had to fact-check it from scratch. The shift he names matters because it changes who is doing the verifying.

The third leg was Sahar Mor at AI Tidbits on his August LinkedIn highlights, an AI-coding-focused roundup centered on DeepWiki, the subagents ecosystem, and Anthropic's own internal Claude Code practices. The throughline of his post is that the tooling layer around AI coding has gotten genuinely useful in the past quarter, with citation-grounded answers and instant repo querying replacing the older pattern of pasting code into a chat window.

Two lighter notes from the same cluster: Marily Nika at AI Product Academy ran two posts in two days, one on Google's Nano Banana image model and one on Opal, Google Labs' no-code mini-app builder. Both are tutorial-style and useful if you build product, less useful as signal.

The take: Narayanan and Mollick are running the same play from different sides. Both are arguing that the operator question has moved from "what can AI do" to "what does the human do when AI does it." The normal-technology frame is the more rigorous version, the wizard frame is the more honest one. Take them together.

Management and Growth: Leaner Teams, Harder Skills

Three operator-focused writers landed pieces this week that, read together, form a single argument about where management is heading. Julie Zhuo at The Looking Glass used the relaunch of her book The Making of a Manager to make the case directly: pure middle-manager roles will decline because AI reduces the coordination work that justified them, but the underlying skill of management (purpose, people, process) becomes more valuable as teams get leaner. The Avengers-versus-Captain-Marvel framing is the line that will get quoted.

Ami Vora at The Hard Parts of Growth ran the most useful tactical piece of the week, on writing a parental leave plan that builds team capacity rather than leaving the team in survival mode. The reframe (treat the leave as a chance to grow your peers and reports into ownership of workstreams you used to hold) is the kind of advice that compounds, and it lands harder in a leaner-team context where coverage has fewer slack hours to absorb.

Kyle Poyar at Growth Unhinged did the cold-water version with "Do you need a GTM engineer?" Poyar expected GTM engineering to be the hottest job of 2025 and instead found 45 job posts in the past month, 128 in the past three, mostly at OpenAI, Ramp, and Webflow. There are more LinkedIn posts about GTM engineering than actual companies hiring for it. His piece is a useful corrective to the role's hype curve.

The grace notes here: Sean Ellis on the most common mistakes founders make when scaling, Paul Stansik at Hello Operator on "You Grow By Trying Stuff" (the case against single-template growth playbooks), and Guillermo Flor at VC Hiring with the weekly VC and startup jobs roundup. Stansik's piece pairs naturally with Poyar's: both are arguing against the comfort of a fixed playbook in a moment when the actual operator skill is calibrated experimentation.

The take: the management writing this week is converging on a single picture. Teams will be smaller, roles will be flatter, and the work that AI cannot do (taste, judgment, knowing which experiment to run next) is the work that gets harder to outsource. Zhuo and Vora are giving you the senior-IC version of the same advice Poyar and Stansik are giving you at the GTM layer. If you only read one, read Vora.

Thinking Traps: Overthinking, Underwriting, and the Cost of Frames

A small but cohesive cluster of writers spent the week on the meta-question of how to think clearly. George Mack at High Agency led with "10 ideas for overthinkers," a long essay arguing that the average person spends nearly eight hours a day mind-wandering (Jan 1 to June 30 of conscious existence, by his math). The high-agency reframe of time as five modes rather than three (past-dwelling, past-rectifying, present, future-worrying, future-building) is the most actionable line. Rectifying the past is 100x better than dwelling on it. Building the future is 100x better than worrying about it.

Steven Schlafman at Where the Road Bends ran the embodied version of the same idea in "The Greed That Never Ends," a confession-style piece about being mentally absent at a playground with his daughter while Bitcoin crashed on his phone. The writing is uncomfortably honest about what it costs to be physically present and mentally elsewhere, and it doubles as the emotional case for everything Mack argues analytically.

dynomight ran the most interesting piece of the cluster with "Dear PendingKetchup," a careful reply to a Substack commenter who flagged the eugenics history embedded in the statistical definition of heritability. The reply takes the critique seriously, works through the technical and political layers separately, and is a model of how to handle a sharp commenter without dismissing them. It is also a quiet argument that the frames we inherit shape what we see even when the math itself is sound.

The lifestyle pendant came from Piera Luisa Gelardi at Noomalooma with a photo-diary of her September, all "Wonder Wanders" and gel-polish scraps and creative-director shoots for her partner's espresso soda brand. The piece is the antithesis of overthinking: presence as discipline, devotion over rigor.

The take: Mack, Schlafman, and dynomight are running variations of the same argument, that the frames we inherit (about time, about money, about heritability) cost us more than the decisions we make inside them. The cheap version is "be present." The hard version is noticing which frame is running you. Mack gives you the analytic case. Schlafman gives you the lived one.

Grace Notes

A handful of pieces did not fit a cluster but earned a mention. Gretchen McCulloch at Lingthusiasm-adjacent Substack announced that her cohost Lauren Gawne has written Gesture: A Slim Guide, an entry-point text for the gesture literature that Gretchen herself wished existed back in 2017 when she was writing the emoji chapter of Because Internet. Addy Osmani at Elevate wrote a deep technical overview of how modern browsers work, covering Chromium's architecture from networking through Blink rendering through V8 and the multi-process sandbox model. Internal Tech Emails surfaced a 2002 Larry Ellison memo on Oracle CRM that reads exactly like a 2025 founder email ("we cannot remain in denial if we talk with our users every day"), plus a draft-versus-final comparison of Steve Jobs's 2007 open letter on third-party iPhone apps. Snaxshot flagged creatine as the next CPG beverage boom.

The honest noise of the week: Upen at Micro SaaS Ideas launched a course library to its 40,000 subscribers, signull ran a short prose poem on the gap between curated aesthetic and actual story ("the cover art never lasts"), Winning Therapy at The Winner's Almanac returned from a three-week absence with a Tesla quote and a maximalist post on the Victorious Attitude, and JJ Chou ran a three-part Mandarin reading-notes series on Nick Maggiulli's The Wealth Ladder.

Three Takeaways from the Week

The clearest signal of the week was that the AI conversation has moved past the GPT-5 disappointment cycle and into a more grounded frame. Narayanan and Mollick are arguing for the same shift from different angles: the question is no longer what AI can do in a benchmark, it is what humans do when AI is doing the first draft. The normal-technology framing is the more useful one to carry into Q4.

The management cluster was the most operator-relevant thread of the week. If you run a team in 2025, the convergent argument from Zhuo, Vora, and Poyar is that headcount is going down, the work that survives the cut is the work that requires judgment, and the playbook for the GTM engineering role got further from reality than the discourse suggested. Plan accordingly.

If you only revisit three pieces from the week, I would suggest Arvind Narayanan's guide to understanding AI as normal technology for the frame, Ami Vora on writing a great parental leave plan for the operator move, and George Mack's "10 ideas for overthinkers" for the personal reset. A thin week, but the pieces that landed were the ones worth slowing down for.