Friday, February 28, 2025 · 2 newsletters

The Sparse Inbox Year

deepseek-and-the-ai-thesis-revision · off-balance-sheet-state-finance · identity-and-inherited-scripts · late-cycle-market-bind · practice-as-discipline

Pulled from roughly 44 newsletters across five publishing weeks. The whole month never crossed double digits in any single week. The first week opened with DeepSeek's R1 rattling the NVIDIA trade, the last week closed with NVIDIA earnings getting punished on a beat. In between, a working historian conceded the AI tools now matter, two writers documented the off-budget plumbing of US and Russian state finance, a triptych of birthday essays asked what version of yourself you are allowed to become, and a value investor finally pivoted to outright shorts. February 2025 was not a year of breaking news cycles. It was a year of sparse inboxes where the pieces that earned a slot had to carry themselves.

The Month in One Sentence

This was the month the AI scaling-laws thesis got its first credible counter, the value-investor playbook ran out of answers, and the writers worth following were the ones not straining for relevance.

Arc: The AI Thesis Gets Stress-Tested in Public

The month's structural arc opened with a single market-moving event and ended with a builder-layer that kept compounding while the trade rerated. Week one was the DeepSeek shock. The Last Bear Standing was the only newsletter in the inbox that engaged seriously with R1's release, and the reading was the calibrated one: the $6 million pre-training figure is misleading, the true buildout cost is closer to $500 million, but the models are real and the scaling-laws thesis that has anchored American AI investment since the OpenAI paper of January 2020 just got a credible counter-example. The market reaction was the loud part. The thesis revision was the quiet part.

Week two named the agent vocabulary while the field was still calling it marketing. Mark Humphries at Generative History wrote the post that should have spread further than it did, conceding that OpenAI's Deep Research outputs "would pass muster in any historical research firm or PhD level course." A working historian, not a tech commentator, doing the frame-shift academics had been deferring for two years. Sahar Mor at AI Tidbits ran the supply-side companion, surfacing Anthropic's Building Effective Agents and the five canonical patterns (prompt chaining, routing, parallelization, orchestrator-workers, evaluator-optimizer) that became the vocabulary the field has been using ever since. The week these were being named, almost nobody flagged it.

Weeks three and four went quiet on AI by volume but the builder layer never slowed. By month-end, Mor was back with his February LinkedIn highlights, flagging Anthropic's counterintuitive note that placing long documents (20K+ tokens) at the top of your prompt, before the query, lifts Claude's accuracy by roughly 30%. Documents first, structured organization via XML, specific query at the end. The single most useful operator tip of the month, dropped into a roundup nobody was reading for breaking news.

Week five closed the arc on the trade side. The Last Bear Standing rang the bell on February 28 with "Max Stupid." NVIDIA beat earnings on Wednesday and got punished with an 8.5% plunge on Thursday, trading back to its DeepSeek lows. The framing line: "when good news is punished, it's a telltale sign that expectations have run ahead of reality." The Four Shorts basket from the prior week was down 21% to 30% in five trading days. The DeepSeek shock and the Max Stupid call were the same arc read at two altitudes. The first one revised the thesis. The second one priced it.

Arc: The Off-Balance-Sheet Read on State Finance

The most rigorous single cluster of the month came in week three, when two writers published long, technical pieces on how governments and central banks fund themselves through mechanisms that do not show up where you would expect them to.

Craig Kennedy at Navigating Russia published the full working draft of "Russia's Hidden War Debt," identifying a previously underreported pillar of Russia's war financing. Alongside the highly scrutinized defense budget, Moscow has been running a parallel off-budget scheme since February 2022, directing major Russian banks to extend preferential loans on state-set terms to businesses providing goods and services for the war. The scheme has driven roughly 36.6 trillion rubles ($446 billion) of anomalous corporate borrowing since mid-2022, has become the main driver of Russia's 10% inflation, and is starting to constrain the Kremlin's war calculus in ways analysts who looked only at the official defense budget missed. The source documentation was unusually deep: the underlying NBER working paper, the Russian enabling legislation, the precedent statute the Duma adopted.

The Last Bear Standing ran the domestic counterpart with "Running on Empty," a sharp look at the Fed's three years of Quantitative Tightening. The argument: the Fed's $2.1 trillion balance-sheet reduction since June 2022 is largely an illusion of accounting, because over the same period usage of the Fed's overnight Reverse Repo Facility has fallen by $2.2 trillion, fully offsetting the headline shrinkage. The monetary base has actually increased. The piece flagged that the RRP overflow tank is now down to $68 billion, the lowest since early 2021, and that as it trends toward zero, QT will for the first time actually begin to tighten by draining real banking liquidity rather than mopping up parked cash.

Read together, these two were doing the same work in two different jurisdictions. The visible balance sheet of state finance is misleading, and the action lives in the off-budget or off-balance-sheet plumbing that conventional analysts undercount. Kennedy on Russia is the more dramatic claim, but the Last Bear piece is the one that matters for the next twelve months of US markets. The cluster was small. It was the most rigorous reading of the month.

Arc: Inherited Scripts, Three Decades In

The most cohesive non-industry thread of the month was a quiet triptych of identity-and-ambition essays in week three, then an echo in week five from a different cluster of writers.

Week three ran the original triptych. Alec McNayr wrote a birthday post built around a Venn diagram about why he keeps making dad-joke videos and performing them at open mics. The piece was ostensibly about choosing a college major decades ago and ostensibly about wondering whether he is, finally, an artist. What it was actually about was the permission an adult gives themselves to make things that might be embarrassing. Liz Tran turned 40 and wrote her manifesto on Life Skill #3, "Fuck It and Be Happy," anchored by the comparison to her own mother at 40 in mini skirts versus herself in sensible shoes and loose pants. Her practice, recapturing childhood satisfaction rather than recapturing her twenties, was the cleanest framing of how to age well in the month. Steven Schlafman ran the third panel with "Do You Have to Grind to Be Great?", sitting with a question Jason Jacobs put to him on a podcast and rejecting the inherited frame that greatness requires being tortured.

Week five echoed the same shape from a writer in a different domain. Schlafman came back with "Slow Enough to Matter," admitting that the Downshift program he built to help ambitious professionals slow down was itself still being driven by the part of him that refused to stop. The wound has a name: "ever-becoming," the relentless urge to stay in motion. Jenny G. Zhang wrote the quieter version with "The truth about writer's block," remembering writing three novellas by age 13 and then losing the voice for six years, having taught herself that writer's block was a thing that happened to writers and that the Muses would return when they felt like it. She now calls this "the worst possible lesson I could have internalized."

The four writers across two weeks were doing the same work: noticing an inherited script (artist as suffering bohemian, mother as harried sacrificer, founder as obsessive grinder, writer as tortured savant) and asking whether you have to keep performing it. The fact that this many independent writers landed on the same shape across the same month with no obvious news peg made it the read of the month worth carrying forward.

Arc: The Value Investor's Bind, Then the Break

The market arc of the month started ambivalent and ended decisive. Week two had The Last Bear Standing running "Framing the Homebuilders," asking whether the post-2020 housing trade was at an inflection. Existing-home supply was frozen by the mortgage lock-in effect, new construction had filled the gap for three years and printed major profits, but homebuilding stocks had given back much of the prior year's gains. The piece sifted six winners from four losers and flagged the "haven't we seen this movie before" question explicitly. A calibrated take on a single sector.

Week four was the value investor's bind out loud. The Stonkstack opened with the line that anyone watching this market had muttered: "What the hell am I supposed to buy now?" AI stocks at 30x sales, IPOs doubling on day one, profitless companies at billion-dollar valuations. The answer was to lean into special situations the way Buffett did in his partnership years, when traditional value plays got crowded out. The Last Bear Standing ran "Four Shorts" in the same week and shifted tone for the first time: the speculative-name profile (negligible revenues, major losses, grand narratives, massive price appreciation) was now actionable on the short side. The rocket had exhausted its fuel. The companion artifact was Fintech Compliance Chronicles on the Capital One / Discover shareholder vote getting 99.8% approval at Capital One representing 85.1% of outstanding shares, an unusually decisive verdict on the largest US card-issuer combination in a generation.

Week five priced the break. "Max Stupid" was the bell. The Four Shorts basket dropped 21% to 30% in five trading days. The momentum junkies who had been chasing 10% daily jumps in Palantir, Robinhood, Hims and Hers, and Applovin were "yacking up their blow-off tops." Bitcoin was "swimming with cement boots." The discount-to-intrinsic-value playbook did not have an answer for late February 2025, and the writers worth following this quarter were the ones admitting it out loud and then doing something about it.

Arc: The Practice of Showing Up

The smallest arc by volume, the most useful by transfer. Week four ran Alec McNayr's post of the month: fifteen months of monthly attendance at The Moth's live storytelling events in Los Angeles. Eleven shows bought, name in the hat every time, five stage slots actually pulled. Seventeen names for ten spots. Forty-two names for ten spots. Thirty-three for ten. The framing he used, paraphrasing Ira Glass, is that he is in the gap between taste and execution, where you can see what good looks like but you cannot yet make it. The discipline is to keep buying the ticket. That is the kind of practice essay that earns its slot.

The institutional companion came from Abby Falik at Taking Flight, opening Flight School applications for the second cohort with the founding Fellows nine months into the year. The pitch was the same instinct at a different scale: the audit on whether a year-old institution is real happens later. The grace-note companion came from Gabby Lord at omglord on journaling as drain-the-brain practice. Three different scales, one argument: the work is in the showing up and the audit happens later. McNayr's 5-of-11 ratio against fields of 17 to 42 names was the version with teeth, because he named the actual odds and refused to dress them up.

The Story of the Month

The story of the month was DeepSeek and the AI thesis revision, with the back-half rerating as the second half of the same arc. The first week opened with the only newsletter in the inbox that took R1 seriously, and the last week closed with NVIDIA earnings getting punished on a beat. Between those bookends, the agent vocabulary got named, the working academics started conceding the tools mattered, the builder layer kept shipping useful primitives, and the speculative names finally broke. The arc was one continuous story.

The case for this as the story rather than the off-balance-sheet finance cluster or the inherited-scripts triptych: every other thread in the month was running adjacent to it. The value investor's bind exists because the AI trade priced perfection into anything with a story. The Capital One / Discover deal got 99.8% shareholder approval because the alternative is being a sub-scale bank in a world where the picks-and-shovels names are eating everything. The off-balance-sheet state-finance read matters because the Fed is now constrained in exactly the conditions an AI-capex boom would otherwise demand cheap money. Even the inherited-scripts triptych is a downstream artifact: when the dominant story of work is being rewritten in real time by a technology nobody fully understands, the writers asking what version of yourself you are allowed to become are answering a question the moment is forcing on them. The story of February 2025 was that the AI thesis stopped being the only credible thesis, and almost everything else worth reading in the month was responding to that.

In Retrospect

The Bear's homebuilder question aged toward "no." Week two flagged the "haven't we seen this movie before" question on homebuilders explicitly, sifting six winners from four losers. The honest update from later in the year is that the answer was closer to no than yes, but the question was the right one to ask. The piece's value was in the framing, not the call. Reading it now, the calibrated refusal to predict was the discipline.

The "Deep Research will pass muster" call was right at the wrong altitude. Mark Humphries' February 9 piece on Deep Research was the frame-shift, but his line about hallucinations being "launch-state artifacts rather than inherent limits" turned out to be more optimistic than the next eighteen months of agent research warranted. The tools did matter. The hallucinations did not vanish. The right read on Humphries now is that he got the demand-side concession right and the supply-side timeline wrong, which is the better mistake of the two to make.

The MAHA-as-NRA argument from Justin Mares looked sharper in February than it does in retrospect. Mares at The Next made the strategic case that health had become a voting issue and that the opportunity was to build "the NRA for health." The frame was correct in shape. The execution evidence over the following months suggests the single-issue coalition mechanics he was modeling do not transfer cleanly from gun politics to health politics, because health is structurally more diffuse and more captured. The argument was right that small organized coalitions punch above their weight. It was wrong about which coalition was ready to be that one.

What to Carry Into Next Month

The sparse-inbox discipline is the thing to carry forward as a reading habit. Every week of February was a week where the news flow did not earn the slot, and the pieces that survived were the ones that knew exactly what they were. McNayr's loss-rate essay survives because he named the odds. Kennedy's Russia paper survives because the source documentation was the work. Schlafman's "Slow Enough to Matter" survives because he admitted the program he built to fix the problem still had the problem. The reading habit to carry into March is to weight pieces by how honest they are about their own limits, not by how much they claim to know.

The AI-trade arc has a second half coming and the writers worth following are already identified. Last Bear Standing called the DeepSeek shock in week one and the speculative-name break in week five. Sahar Mor caught the agent vocabulary the week it was being named. Humphries caught the academic concession. Ben James, announcing his move to Substack with Ben by Fax in the final week, is the writer to add to the list, because "data centers will be powered by solar and gas" is the cleanest piece of energy-realism writing of the month and the hybrid-generation answer is going to look obvious in eighteen months. March's reading list writes itself from this group.

If you only read three pieces from February, I would suggest The Last Bear Standing on the DeepSeek moment for the cleanest read on the one market-moving event of the month, Craig Kennedy's Russia's Hidden War Debt for the deepest single piece of original research of the month, and Alec McNayr on showing up at The Moth for the practice essay that transfers to any creative discipline where the loss rate is structurally high. The month was sparse. The pieces that survive it are the ones that knew exactly what they were.