Week 21 · 2025-05-19 → 2025-05-25 · 10 newsletters

Design, Layoffs, And Loan Math

ai-labor-reckoning · design-led-ai-hardware · finance-under-the-hood · google-research-to-reality

Ten emails across seven days. A sparse week, but a coherent one: the inbox kept circling back to what happens when AI hits an actual P&L, an actual product roadmap, and an actual balance sheet. Microsoft cut 7,000 people on Monday, OpenAI bought Jony Ive's startup on Thursday, and in between, two finance writers picked apart what looks cheap and what looks load-bearing. Google I/O happened. A few writers reflected. That was the week.

AI Labor Reckoning: When the Layoff List Has Names On It

Jacob Voytko at Client/Server led the week with the most uncomfortable read: Microsoft announced a 3% workforce reduction, roughly 7,000 people, with Bloomberg reporting 40% of the cut were software engineers. Voytko did the back-of-envelope math (Microsoft listed over 100,000 developers against 181,000 employees in 2021, so engineers were actually spared relative to other departments) but the part that landed was the named list. The "faster CPython" project was wiped out, including 3 core Python developers. A director of AI was let go. At least one person responsible for Microsoft's 10x TypeScript compilation speedup was cut. Microsoft itself said the cuts were made without regard to performance.

Ethan Mollick at One Useful Thing ran the demand-side companion piece, "Making AI Work: Leadership, Lab, and Crowd." His four-fact framing was clean: AI boosts work performance, a large percentage of people are using AI at work (40% of American workers as of April 2025, up from 30% in December 2024), the transformational gains available exceed what most realize, and yet companies are reporting only small to moderate gains. The gap between individual productivity and enterprise capture is the operator problem of the year, and Mollick is the writer doing the most disciplined work on naming it.

The take: read Voytko and Mollick together and the picture sharpens. Companies cannot find the productivity gains in their P&L, so they cut headcount to manufacture them, and the cuts hit the engineers who were actually delivering improvements. Microsoft laid off the person who made everyone's TypeScript compile 10x faster. That is not a productivity story. That is a story about executives who do not know what their own organization produces, making decisions about who produces it.

Design-Led AI Hardware: A $6.5B Teaser, And A $500K Counter-Bet

The week's biggest single story was OpenAI's $6.5B all-stock acquisition of io, the AI hardware startup co-founded by Jony Ive, Sam Altman, and former Apple design leads Evans Hankey, Tang Tan, and Scott Cannon. Carly Ayres at Good Graf ran the sharpest annotation of the announcement, calling it "less a take, more an annotated moodboard." Her observation that the announcement was a teaser rather than a product launch (something Apple in the Jony and Steve era never did) is the right frame. The nine-minute video of Jony and Sam walking through soft San Francisco light said very little about what they are building, which is itself the point: OpenAI is buying the design-led narrative before it has the product to attach it to.

The companion piece, also Ayres, came earlier in the week from AI Residency (AIR), the NYC accelerator she helps run, opening its summer cohort application for "early teams with taste: founders designing for emotion, intuition, and cultural impact, not just infrastructure or optimization." AIR is putting $500K behind that bet, which is small money against OpenAI's $6.5B, but the thesis is the same one: that the next wave of consumer AI is a taste question, not a model question.

The take: Ayres is running both sides of the same argument in the same week, and she is right to. If OpenAI's io acquisition lands, it validates a thesis the design-led AI community has been pushing for two years: that the bottleneck is not the model, it is the interface around the model. If io ships nothing recognizable in 2026, the thesis takes a hit and the AIR cohort thesis takes a hit with it. Either way, the next twelve months will tell us whether design-led AI is a real category or a $6.5B teaser.

Finance Under The Hood: Two Writers, Two Balance Sheets

The Stonkstack and The Last Bear Standing both ran detailed balance-sheet pieces this week, and together they form a useful pair on what "cheap" actually means in 2025. Stonkstack's "What's In My Deep Value Portfolio? #3" pushed back on the post-tariff narrative that US equities are full of bargains, noting that across managed accounts roughly three-quarters of the equity exposure is non-US, "as honestly, there's really not that many mouth-watering opportunities in the United States." The line about Substack writers promising "a brand-new 'hidden gem' every two business days" being mostly value traps or mediocre deals selected for their ability to lure subscribers was the editorial cut of the week.

Daniel Saedi writing at The Last Bear Standing did the harder work on SoFi in "So-Fried." His argument: SoFi has quietly transformed from a fintech originator into a regulated specialty consumer lender, with an OCC bank charter funding aggressive growth in unsecured credit card consolidation and personal loans, while continuing to book profits as if it were still an originator. The damning specifics: nearly a third of SoFi's tangible equity capital has come from cumulative write-ups of its loans, even as credit performance has materially declined. If SoFi valued its loans at cost like its peers, Saedi argues, it would likely trip bank capitalization requirements and prompt immediate corrective action.

A third finance note came from Zarik Khan at Fintech Compliance Chronicles, whose "Voices of Discover" was a personal reflection on the now-finalized Discover-Capital One deal after his eight and a half years at Discover. Less an analysis than a memoir, but a useful reminder that integration work is where the real story starts, not where it ends.

The take: Stonkstack and Saedi are doing the kind of work most finance Substacks do not. Stonkstack is naming the bezzle of the bargain-hunting genre. Saedi is doing the accounting forensics that take a long weekend to write and that nobody else in his SoFi coverage is doing. Both pieces are reminders that the value of independent finance writing right now is not the stock picks, it is the willingness to look at the part of the disclosure nobody is reading.

Google I/O 2025: Research To Reality

Sahar Mor at AI Tidbits ran the cleanest readout of Google I/O 2025 in "Research to Reality." His framing: the AI community has spent recent years viewing Google as trailing OpenAI and Anthropic, but at this year's I/O "everything finally clicked." Google has moved from research output to product reality by leveraging its distribution channels (Search, Workspace, Android, smart TVs, glasses, phones) in a way that no competitor can match. Mor compared the energy at the press tent to OpenAI's inaugural DevDay, which is the kind of comparison that means something coming from him.

The take: the Google story all year has been "Google is behind." The Google story now is "Google has the distribution and is starting to use it." If you read Mor's piece alongside the OpenAI/io news in the same week, the symmetry is hard to miss: OpenAI is buying its way into the hardware distribution problem that Google solved twenty years ago. That asymmetry is going to matter more in 2026 than the model leaderboards will.

Three Takeaways from the Week

The Microsoft layoffs were the labor story of the week, but the deeper story is the gap Mollick named: individual workers report large AI productivity gains, companies report small ones, and the reconciliation is happening through headcount cuts that are not actually targeting unproductive people. Watch this gap. It is the question that decides whether 2026 is the year AI capture works or the year it fails.

The OpenAI/io acquisition and Google I/O happened in the same week, and the contrast is the through-line. OpenAI is paying $6.5B for design talent to invent a hardware category. Google already has the hardware categories and is finally shipping the AI into them. Both bets could pay off, but they are different bets, and the writers covering only one side are missing the symmetry.

If you only revisit three pieces from the week, I would suggest Jacob Voytko on the Microsoft layoffs for the most honest read on how the cuts actually looked from the bottom up, Carly Ayres on the OpenAI and io announcement for the sharpest annotation of the week's biggest deal, and Daniel Saedi on SoFi for the kind of balance-sheet forensics that justify the price of a paid Substack on their own.