Week 32 · 2025-08-04 → 2025-08-10 · 11 newsletters
GPT-5 Lands in a Quiet Inbox
gpt-5-and-the-leverage-question · building-quietly-vs-vibe-coding · markets-getting-stupid
The first full week of August, eleven emails across seven days. Sparse by any standard, but the timing was loud: GPT-5 shipped midweek and gave the inbox its center of gravity. The rest of the week split between two long-running arguments (how founders should think about leverage now that code is the bottleneck, and whether the markets have lost their minds) plus a few quieter pieces on building things that last. No politics, no macro, no big policy threads. Just operators trying to figure out what GPT-5 actually changes.
GPT-5 and the Leverage Question: Capital, Labour, Code
The week's center was Ethan Mollick at One Useful Thing on GPT-5, which he had early access to and which he framed unambiguously as a big deal. The opening move was the demonstration: he asked GPT-5 to do "something very dramatic" inside the intro paragraph, and it produced a passage where the first word of each sentence spells out "This is a Big Deal," each sentence is exactly one word longer than the last, most words in each sentence start with the same letter, and the whole thing reads as coherent prose with a style. The point is not the trick itself, the point is that GPT-5 planned, executed, and held the multi-constraint problem in working memory long enough to ship it in one pass. Mollick's phrase for what the model now does is "it just does stuff." That is the line to remember from the launch.
The companion piece, written the same day by Kerman Kohli, did not mention GPT-5 by name but framed the implication. The pathway to leverage, he argued, has collapsed: Capital then Labour then Code then Leverage became, in the Sonnet 3.5 plus Cursor era, just Code then Leverage. If you can prompt correctly, you can generate labour leverage independent of human labour or dollars. He extended the historical arc: in ancient Egypt you had to be born a Pharaoh, a few hundred years ago you had to be born a King, more recently you had to raise venture capital to pay salaries. Now you type the right keys. The unstated argument running underneath GPT-5 week is that the compute ceiling, not the model capability ceiling, is now what bounds individual leverage. Kohli's title was "We don't have enough compute," and he meant it literally.
Brianna Zuniga at Circular Architect ran the cognition counter-argument with a deliberately facetious title ("what you're all getting wrong about ai and human cognition") aimed at the five-plus pieces she had seen that week arguing AI would atrophy human thought. Her pushback was the right one: enfeeblement is not inevitable, it is a design choice, and the real question is whether tools sharpen cognition or quietly hollow it out. She cited Neil Cybart of Above Avalon on the fine line between augmentation and enfeeblement. The piece earns its slot because it refuses the easy fatalism without dismissing the concern.
The take: GPT-5 is the inflection event of the week, but Kohli and Zuniga are the writers who got at what it actually means. Mollick described the capability. Kohli described what changes for solo operators when capability stops being the bottleneck. Zuniga described what the operator still has to decide for themselves. Read in that order, the three pieces are the week.
Building Quietly vs. Vibe-Coding: Two Models of What Lasts
A second through-line, smaller but cohesive, ran through three pieces about how founders should actually build right now. Carly Ayres at good graf used Figma's IPO as the frame for "The designer investor era," and the line that did the work was the contrast: in an era of vibe-coded apps launched over a weekend with AI prompts and momentum theater, Figma offers a counter-narrative. Founded 2012, four years before public debut, seven more to profitability, a $20B Adobe acquisition that fell through, $411M raised, thirteen-year overnight success. Built slowly, shipped carefully, meant to last. Ayres' point is that designers with capital, context, and a bias toward building are about to shape what gets funded, but the underlying argument is sharper: enduring companies are rarely built on vibes alone.
Sean Ellis pointed at the article he recommends most to founders, Paul Graham's "Do Things that Don't Scale," and the framing rhymes with Ayres. The pre-product-market-fit work is unglamorous, manual, and most of it does not look like leverage in the moment. Ellis has been pointing founders at this essay since his early Y Combinator days with Xobni and Dropbox, which makes the recommendation feel less like a take and more like a settled position after twenty years of pattern-matching.
Alex Konrad at The General Partnership hosted Dan Shipper of Every for a conversation on building modern media companies that lands in the same territory. The points worth pulling: being weird is often a competitive advantage, thinking like a product person rather than a writer shapes everything Every publishes, both Upstarts and Every resisted the pressure to "build the platform" too soon, your business model is a creative decision, and Every runs multiple AI products plus a daily newsletter with fifteen people. The Claude, Cora, and ChatGPT references in their creative process are concrete, not aspirational.
The counterweight came from Upen at Microsaasidea cataloging the vibe-coded end of the spectrum: TypeThinkAI at $200 MRR, How To Convert at $10K in eight weeks, Coolify at $15K/month, Base44 bootstrapped to $189K/month in six months before Wix bought it, CodeSpring at $15K/month, Shhots AI at $15K/month. The aggregate picture is real, but the survival curve underneath those numbers is the part the post does not address. signull titled the week's bluntest piece "stop playing games you can't win," with the Federer-not-grinding-on-clay frame: some surfaces are not yours, some games are unwinnable by design, some markets are meat grinders for people like you. Short, punchy, and the right counterweight to the micro-SaaS aggregator.
The take: Ayres, Ellis, and Konrad-on-Shipper are running the same argument in three keys, that the slow build is the durable one and the vibe-coded weekend launch is mostly noise. signull is the cold-water version. Upen is the survivorship-bias version. Read them as a set and the through-line is clear: leverage from code is real (Kohli was right earlier in the week), but leverage from code does not absolve you of choosing a game you can actually win.
Markets Getting Stupid: The Bitcoin Treasury Trade
The week's one piece of finance writing was The Stonkstack on a Financial Times article titled "Why struggling companies are loading up on bitcoin," which the author predicted will one day be used in case studies the way Dot-Com Bubble texts are read now. The argument is that Michael Saylor's playbook of raising cash from investors and gambling it on bitcoin is being adopted by genuinely struggling companies as a turnaround strategy in lieu of, say, improving the underlying business. The piece is profane and unmeasured and probably right about the case-study point. Stonkstack pulled out Howard Schilit's "Financial Shenanigans," page 33, where Saylor's original gambit is documented, to make the historical case that this pattern has been described in textbooks for two decades.
The companion piece, oblique but related, was Internal Tech Emails on Mark Zuckerberg's April 2012 internal note about getting the Instagram deal done at 1.3% or 1.4% under "extreme time pressure." Read alongside the Stonkstack piece, the throughline is that the deals that end up in case studies are the ones the participants knew at the time were either generational steals or generational mistakes. The Instagram acquisition has aged as the steal. The bitcoin treasury trade is, per Stonkstack, aging the other direction.
The take: the only finance writing of the week was a profanity-laden FT-citation, and the only deal-history piece was a thirteen-year-old email about Instagram. That is what a sparse week looks like. The signal is that nobody serious had much to say about markets this week, and the people who did were either flagging the absurd or excavating the historical.
Three Takeaways from the Week
GPT-5 launched midweek and the framing that actually mattered was not the capability description but the leverage implication. Mollick on the model itself is the technical read. Kohli on Capital-Labour-Code collapsing to just Code is the operator read. If you have to pick one frame for what changed this week, it is Kohli's: the bottleneck moved, and the new bottleneck is compute, not capability.
The second-largest cluster was the slow-build versus vibe-code argument, with Ayres on Figma and Ellis on PG's essay and signull on unwinnable games all making the same point from different angles. The micro-SaaS aggregator running underneath them is the noise, not the signal. Pick the game before you pick the tools.
If you only revisit three pieces from the week, I would suggest Ethan Mollick's GPT-5 first impressions for the cleanest read on what the model now does, Kerman Kohli on compute and leverage for the cleanest read on what that means for solo operators, and Carly Ayres on the designer investor era for the cleanest read on what kind of company actually lasts when everyone else is shipping in a weekend. The week was quiet. The pieces that earned their slot are the ones that took the launch seriously without confusing capability with strategy.