The layoffs aren't AI — and the source code isn't ransomware

00:00:04 A YouTube short titled "Polite Hackers" by The PrimeTime grabbed my attention first, mostly because of how it was framed. In a thirty-seven-second clip, an actor in a casual office setting leans into the camera and reads from a list: No lowball offers will be accepted.

00:00:28 Everything for the main platform is there and I am very happy to send samples to interested buyers to verify the absolute authenticity. There's around 4,000 repos of private code." We do not care about extorting GitHub. One buyer and we shred the data on our end.

00:00:50 It looks like our retirement is soon. So, if no buyer is found, we will leak for free." What matters here isn't the breach itself—breaches of this scale happen and fade in a news cycle—but the framing. The actor treats the stolen data exactly like a marketplace listing.

00:01:18 There are no ransom notes or demands, just inventory. One buyer triggers a payout and they vanish; if no buyer comes forward, the leak goes out anyway. That posture signals how the underground economy is shifting. Ransomware has been dominant because it creates urgency and negotiation leverage.

00:01:39 This approach rejects that dynamic entirely. It wants to be transactional and fast, which tells you something about how the market is maturing: not toward more chaos, but toward streamlined customer service. I haven't found confirmation yet of whether these repos actually belong to GitHub or if this is a social engineering pitch to hook buyers first.

00:02:04 Either way, treating source code as a flat commodity rather than as leverage is what lingers.

00:02:11 Arvind Narayanan and Sayash Kapoor posted a piece on their Normal Tech blog titled "Why AI hasn't replaced software engineers, and won't." Their argument is straightforward and data-backed. They looked at three major layoff announcements this year that were widely framed as AI-driven: Block's 4,000 cuts (Jack Dorsey cited AI enabling "smaller and flatter teams"), Snap's 1,000 layoffs (Evan Spiegel said AI generated 65% of new code), and Intuit's 3,000 reductions (paired with Anthropic and OpenAI deals).

00:02:48 In every case, the underlying driver was different. Block's cuts came from post-pandemic over-expansion and financial pressure—a Cash App data scientist posted that AI produced "very limited gains in productivity" even as it was pushed aggressively. Snap's layoffs followed activist investor pressure rather than coding displacement; cutting 150 AR division roles doesn't look like the pattern you'd see if programming jobs were actually being automated away.

00:03:21 Intuit's CEO explicitly denied the AI connection. Narayanan and Kapoor point to New York State's WARN Act disclosure requirement, which added an AI checkbox in March 2025. In its first full year, more than 160 companies filed WARN notices. Not one checked the AI box.

00:03:40 One company—Nespresso—did. That accounts for about two-tenths of a percent of laid-off workers in New York State during that period. When there are legal consequences to lying, almost no company checks the box. A Forrester survey found that nine out of ten companies preparing AI-driven layoffs haven't built a mature, vetted tool to replace those workers.

00:04:06 A Federal Reserve paper adds that employment growth slowed by roughly three percentage points annually after ChatGPT's debut—though the methodology misses self-employment, and other studies suggest AI actually lowers barriers to starting a business. What this data shows is that the layoff narratives function primarily as stakeholder management exercises.

00:04:32 Fifty-nine percent of U.S. hiring managers admitted they blame AI for layoffs because it "plays better with stakeholders than citing financial constraints." Which excuse companies prefer matters more than what the technology itself can do. The essay also flags an indirect mechanism: some companies that sell AI, like IBM and SAP, reallocate headcount from legacy functions to their fastest-growing product lines.

00:05:02 That's ordinary corporate restructuring around a revenue opportunity, not technology displacing workers.

00:05:09 Another angle on where AI capital is actually flowing came out in the same reporting this week. OpenAI filed confidentially to go public, but what stands out about their product trajectory is what they're *not* doing. They shut down Sora—which hit one million downloads in five days after its September launch.

00:05:31 They killed Instant Checkout. Their chief revenue officer says they're at a "tipping point" in enterprise adoption, and their CFO projected in March that enterprise would make up 40% of revenue by year-end. Meanwhile, Apple just wrapped WWDC with new features like a standalone Siri app, camera AI, and email integration.

00:05:54 Google I/O last month showcased Gemini Spark, smart glasses, and video tools that let users "change what's happening" in a clip. Both companies are subsidizing consumer AI because they have the capital to absorb the cost: Apple has 2.5 billion active devices; Google has seven products serving over two billion monthly users each.

00:06:18 Gartner analyst Kjell Carlsson put it plainly for Apple: "I can give this away for free, because I'll make it up on the iPhones or iCloud subscription they'll be buying." On OpenAI's side, former Microsoft business intelligence lead Rob Collie said: "That's where we make profit.

00:06:38 That's where productivity is worth paying for." They also acquired Tomoro, an AI consulting firm with 150 deployment specialists. The divergence shows up clearly in their product lines. OpenAI abandoned consumer features that actually worked because they needed enterprise revenue to survive the pivot.

00:07:12 Apple and Google are building those same kinds of features knowing they'll subsidize them indefinitely because the real value lies in keeping users inside their ecosystems. The archive shows OpenAI's ChatGPT went viral with consumers before the company had to abandon that model to secure funding.

00:07:34 Meanwhile, consumer-first companies still back these features, while enterprise-focused models can't sustain their trajectory on consumer revenue alone. This subsidy gap dictates which products ship and which ones get shuttered like Sora despite massive adoption numbers.

00:07:53 Between the layoff essays and the enterprise pivot stories sits something that actually shifts a needle for the outside world. IEEE Spectrum published a piece on Isomorphic Labs—Google DeepMind's drug-discovery spinoff—and their new Isomorphic Drug Design Engine.

00:08:12 What makes it stand out is that it's one of the few AI-in-science stories backed by a technical report with concrete validation. IsoDDE successfully predicted a never-before-observed "cryptic pocket" on the surface of a protein called cereblon. Cryptic pockets are binding sites that don't exist in the unbound state; the protein doesn't have a cavity there until the right ligand binds to open it.

00:08:40 You need the exact key to unlock the lock. Isomorphic asked one question: can IsoDDE find this pocket using only the protein sequence as input? The model found it. Then they asked whether the system could place both the orthosteric ligand at the known binding site and the allosteric ligand at the cryptic pocket in exactly the right locations.

00:09:04 It did. From sequence alone, without any prior knowledge of the pocket's existence, the model mapped something entirely unmapped. Unlike AlphaFold, which predicts known structures, this one works blind. The company raised $2.1 billion, has partners with Novartis and Eli Lilly, and their technical report describes three endpoints: structure prediction, pocket identification, and binding affinity prediction.

00:09:33 They've also validated the system on antibodies, molecular glues, and peptides—not just small molecules. A detail from group leader Adrian Stecuła cuts through the press releases claiming solved problems: It does take, we believe, a unified system such as IsoDDE with a plethora of other endpoints to really model these systems."

00:10:14 Max Jaderberg, Isomorphic's president, framed it as part of an agentic workflow future — "AI systems generating hypotheses, testing hypotheses, and analyzing results" — but the technical report stands on its own as a testable claim rather than a press release about general-purpose agents.

00:10:34 The last item shows where the race has shifted next: both Anthropic and OpenAI are expanding in London. Anthropic hit a $965 billion valuation in its May funding round, while OpenAI came in at $852 billion in March. Both are filing for confidential IPOs—Anthropic first, by about a week—but neither is building new research labs in the U.K.

00:10:57 They're opening offices to be closer to enterprise buyers. The signal here is about distribution and trust, not capability. The models are being developed where the talent already exists. London has become the deployment hub because enterprise clients prioritize data sovereignty, compliance regimes, and local legal jurisdiction.

00:11:20 Anthropic beat OpenAI to the confidential IPO filing, which adds competitive weight even though both companies have been racing on enterprise deals rather than consumer metrics for months now. What stands out is how wide the gap has become between AI narratives and actual work.

00:11:39 Isomorphic Labs can predict a previously unobserved protein pocket from sequence alone using AI. OpenAI shut down Sora despite one million downloads in five days because enterprise customers don't pay for video generation. Companies tell regulators AI caused layoffs but refuse to check the box when it matters.

00:12:00 The archive shows the gap between the public AI narrative and the actual work is widening, not narrowing. Real progress is happening quietly—inside technical reports with testable claims, inside enterprise sales cycles, inside WARN filings where lying carries consequences.

00:12:19 The public narrative keeps running on a loop about replacement and disruption because those stories are easier to sell than deployment and compliance. Next week, we'll see whether the IPO filings proceed and if the NY DOL follows up on that single Nespresso checkbox—or if the disclosure mechanism actually holds water.

00:12:41 — Seln.