Week 17 · 2025-04-21 → 2025-04-27 · 11 newsletters

Driver Or Passenger

ai-agency-and-quality · operator-philosophy · fintech-on-the-ground

Thirteen emails across seven days. A sparse week, but a thematically coherent one: three writers asking the same question about AI from three different angles, three writers running variations on operator philosophy, and one fintech correspondent filing two posts that bracketed the week. Plus the Pride and Prejudice anniversary essay and a linguist's summer travelogue, because the inbox is the inbox. No macro thread, no breaking news, no need to invent one.

AI: Driver or Passenger, and What Counts as Quality Now

Three writers approached the AI conversation from three different elevations this week, and together they read like a single argument about agency. Rob Thomas at The Mentor ran the highest-altitude version with "The Mentor #57: AI", framing it as a driver-versus-passenger choice. His baseline expectation, that before hiring a person you should have to prove AI cannot do the job, is the kind of line that either lands as bracing or as glib depending on where you sit. The post also includes the sharper claim that the rules are still forming and the people who experiment now will be the ones charting the course. It is not a new observation, but it is the right one for a moment when most of the conversation has moved to product reviews.

Julie Zhuo at The Looking Glass, co-writing with Vercel's Guillermo Rauch, ran the most interesting piece of the week in "The AI Quality Coup". The argument hangs on the Studio Ghibli moment from a month earlier: ChatGPT's 4o image launch did not make Ghibli's art accessible, it commoditized the style. The deeper claim is that quality work has always been about hovering at the precipice's edge, and that the new precipice has to be redrawn. The line that stuck: "art is not style." If you are building anything where craft is the moat, this is the post to revisit when the next image model launches.

Sahar Mor at AI Tidbits ran the operator-level version with "Sahar's Coding with AI guide", the first in his AI Coding Series. His thesis is that in a vibe-coding world where Cursor, Windsurf, Cline, and Claude Code can autocomplete a whole feature, speed is no longer the bottleneck. Clarity is. The single rule he keeps returning to: treat your AI coding agent like a highly capable human pair programmer, and a successful session is 80% planning, 20% execution. The Karpathy reference to vibe coding shows up early and gets immediately reframed as something that has to be wrestled into discipline. That is the right move.

The take: read together, Thomas, Zhuo, and Mor are making the same argument from different angles. The opportunity is real, the tools commoditize style before they commoditize taste, and the operators who get value are the ones who treat the agent like a colleague who needs clarity rather than a magic box. If you only read one, read Zhuo, because the Ghibli frame is the one that will keep paying out as the next wave of generative tools ships.

Operator Philosophy: Focus, Mentors, and the Zoho Counterexample

A surprisingly cohesive cluster of operator writing landed this week, mostly arguing the same point from different chairs. Shruti Gandhi at Array VC ran "Zoho's Revolutionary Path", a profile of Sridhar Vembu's $1B-revenue, 12,000-person, 88%-founder-owned company built entirely without venture capital. The numbers are doing the rhetorical work: 80 million users across 180 countries, nearly 20% of engineers without formal degrees, recruited straight out of high school through Zoho Schools. The COVID counterexample is the cleanest part of the piece: while venture-backed competitors did mass layoffs, Zoho accelerated hiring and launched free tools. The bootstrapped model removed the pressure to optimize for anyone except the customer. It is not a new story, but the data points are worth keeping in your back pocket the next time someone tells you a software company has to take outside money.

Winning Therapy at The Winner's Almanac ran two posts that read as bookends. "How To Develop F*ck You Focus" on Friday made the case that extreme outcomes are the product of extreme lifestyles, with the Alexander the Great anchor (one goal, one book, an annotated Iliad under the pillow for over a decade) doing the heavy lifting. The framing of focus as building an iron fortress with a 10-foot moat and Persian longbowmen on the walls is on-brand maximalism. Then Sunday's "On Mentors, The Competitive Spirit, and Diabetes of The Mind" included the Derek Sivers mentor protocol that was the single sharpest paragraph of the week: write the dilemma, predict what each mentor would say, update your write-up, predict again, realize you no longer need to send it. The mentors do not know they are your mentors. That is a tool, not an aphorism.

Jenny G. Zhang ran "Should I write a book? Should you?", which sits on the other end of the operator spectrum: a working media writer thinking out loud about whether to take a literary agent's call. The honest part is the description of how nonfiction books get made (viral essay leads to agent inquiry, hybrid magazine-piece concept gets stretched to 200 pages, often badly). The harder part is admitting that the book she actually wants to write is a novel. The post is the rare piece of writing about writing that does not feel self-indulgent, because the diagnostic on the magazine-piece-as-book industry is genuinely useful for anyone evaluating a similar offer.

The take: Vembu, the Sivers mentor protocol, and Zhang's honesty about which book she actually wants to write all land in the same place. The operators who get to make decisions on their own terms are the ones who have removed external pressure (capital, status, the wrong-shaped opportunity) from the equation. The Persian longbowmen are optional.

Fintech On the Ground: Two Dispatches from Zarik Khan

Zarik Khan at Fintech Compliance Chronicles filed the two most grounded pieces of the week. The midweek post was "Podcast #8 with Prashant Fuloria, CEO Fundbox", a long-form sit-down with the Fundbox CEO covering the working-capital platform's revolving credit line for small businesses (up to $100,000, integrations with QuickBooks, FreshBooks, Xero, Indeed) and Fuloria's path through Yahoo, Google, Facebook, Flurry. The Yahoo detail is the one to flag: he ran a 1300-person engineering and product team driving a $5B-a-year advertising business before the Fundbox move. That is a useful data point on how senior the talent in fintech gets when it commits.

The Saturday post was "NY Fintech Week 2025 Recap", with the news inside the recap being that Rise NYC, the fintech accelerator and hub that hosted the monthly NYC Fintech Coffee event, is closing at the end of May after years as a community anchor. The Compliance and Pastries breakfast at The Smith Nomad, co-hosted with Dmitry Gritskevich of ComplyCo, brought together about 20 founders, lawyers, and compliance professionals. The cleanest measure Khan offers for whether these events work, how many last year's attendees came back this year, is the kind of metric every conference organizer should steal.

Adjacent to this, Jacob Voytko at Client Server ran "How did Google's illegal ad monopoly work?", a plain-language summary of the recent ruling that found Google operating an illegal monopoly across publisher ad servers, buy-side exchanges, and advertiser networks like Adwords. The throughline is that Google's ad server was supposed to act in the publisher's interest (the New York Times in his worked example) and instead repeatedly favored Google's own exchange. The post is what good explainer journalism looks like when the writer disclaims any expertise upfront and then does the work anyway.

The take: Khan's two posts plus Voytko's adtech explainer are the closest thing this week had to an industry signal. Rise NYC closing is the kind of soft news that matters for a city's fintech ecosystem in ways that will not show up in funding databases. The Google ruling will reshape the publisher economics that touch every newsletter in this inbox eventually.

Grace Notes

Jack Butcher at Visualize Value announced the VV app in "VVIP", pitched explicitly as "an opinionated response to infinite doom scrolling," with iOS, widget, web, and Chrome plugin, a free tier of one daily idea, and a $50-a-year pre-launch lock-in price. The line worth borrowing: "to open on purpose, not on impulse." Gabby Lord at Omglord ran "Design and Mr. Darcy are my passion" marking the 20-year anniversary of the 2005 Pride and Prejudice, a roundup of design-and-illustration tributes to the film's lighting, longing, and Lydia (unhinged) energy. Leah Velleman writing in for Gretchen McCulloch filed "July, August, and September 2024" recounting a summer of comparing four language-learning strategies across four languages, plus the Spanish translation launch of Because Internet in Madrid (arroba is the Spanish word for the @ sign, so the localized title literally means @language).

Three Takeaways from the Week

The week's AI conversation has already moved past "can it do the thing" to "what counts as quality once everyone has the thing." Julie Zhuo and Guillermo Rauch's argument that ChatGPT's 4o launch commoditized Studio Ghibli's style without making its art accessible is the frame to carry forward. The next image model will not change this; it will sharpen it.

The single most useful tactical paragraph of the week was Derek Sivers's mentor protocol, embedded inside Winning Therapy's Sunday post. The protocol replaces a meeting with a written prediction exercise and ends, usually, in not needing the meeting. That is the kind of tool that pays off the second time you use it.

If you only revisit three pieces from the week, I would suggest Julie Zhuo's "The AI Quality Coup" for the cleanest frame on what changes when style becomes a commodity, Shruti Gandhi on Zoho for the most concrete counterexample to the venture-default playbook, and Sahar Mor's Coding with AI guide for the 80/20 planning rule that will save you the most hours over the next month.