Product1 distinct publisher3 min readPublished
A New York Times investigation traces Microsoft, Google and OpenAI's hold on US schools to professional development rather than procurement. The pharma comparison is the load-bearing part.
The Product Desk · Product desk
Compiled by The Product DeskSomething wrong?How this is made
Detailing works because the person being persuaded is not the person paying, and because the persuasion is filed as education. Natasha Singer's comparison is to the drug-company seminars for doctors she covered years earlier, and the Georgetown pharmacologist she quotes, Adriane Fugh-Berman, states the mechanism without decoration: industry funds what helps industry [4]. A trained teacher is a better distribution asset than a signed district contract. She picks the tool, raises the requisition, defends it in the staffroom, and does the last mile of implementation at no cost to the vendor.
The sharpest evidence in the piece is an absence. Apple gives away an Advanced Placement computer science curriculum built on Swift, its iPhone language, and Microsoft offers a competing AP Computer Science Principles course with Minecraft Education [9]. There is no Raytheon AP physics course and no Pfizer biology course [10]. Brad Smith's answer is that public schools looked ill-equipped to teach modern subjects and private philanthropy filled the gap, and that this is not how public education should work [11]. The second half of that is the sentence a regulator copies out.
Set the sums side by side. Google agreed in 2019 to pay $170m over regulator complaints that YouTube had illegally harvested children's data, and millions of schoolchildren have since taken its Be Internet Awesome privacy lessons [15]. That settlement, covering one data channel, was roughly seven times the size of the fund now paying to train American teachers on chatbots [18]. Enforcement has priced student data. It has not priced the teacher.
What capture looks like on paper is duller than the workshops. After the Gone Google campaign, schools across nine states published near-identical handbook lines about Chromebooks empowering students to reach their potential [14]. That is vendor prose arriving in district policy without a bid, an evaluation, or a named author. Singer reached her conclusion about a handful of firms shaping nearly every step of the supply chain from more than 100 interviews and hundreds of school, government and corporate documents [3], and the documents are where the channel is visible.
Randi Weingarten conceded the obvious point to Singer, that some of the companies her union is working with are in it for profit and that profit and public education are different goals [6][7]. For anyone selling into institutions, the operational read is narrow and unhappy: your enablement programme is your distribution channel, and it is the part of your go-to-market with the least documentation. If a state attorney general applies the pharma frame, the first request will be for the training deck, the trainer list, and the accounting treatment. Procurement keeps those records. Philanthropy budgets usually do not.
Ranked by verification strength, evidence, and original report placement.
Singer interviewed more than 100 education and tech industry figures and reviewed hundreds of school, government and corporate documents, concluding that a handful of giants have influenced nearly every step of the education supply chain.
The workshop reminded Singer of seminars organised by drug companies for doctors that she had covered years earlier; Adriane Fugh-Berman, a Georgetown pharmacology professor who studies drug industry influence on medicine, told her: "Industry funds what helps industry."
Lois Weiner, education professor emerita at New Jersey City University and a retiree member of the United Federation of Teachers, argued the national union was selling out, telling Singer: "This isn't an alliance. We aren't sitting at the table. We're the meal."
Eric Schmidt, then Google's executive chairman, told an audience in Scotland in 2011 that British schools were failing to teach computer science; Microsoft issued a similar admonition to the US a year later citing an engineer shortage; and the message took off in 2013 when the nonprofit Code.org posted a video starring Bill Gates and Mark Zuckerberg.
On a humid afternoon in the summer of 2025, Microsoft representatives presented Copilot to about 200 teachers in a downtown Manhattan conference hall, having them practise writing emails to parents and customising lesson plans, and screened an explainer video from Microsoft's Minecraft Education division.
Natasha Singer, who has covered school technology for the New York Times for more than a decade, was in the room and published her findings in the paper on Sunday; the article is an adapted excerpt from her book Coding Kids.
Follow any of these and your For You feed starts watching them — no settings page required.
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Named, quoted and quantified — but relayed through one publisher
The substance is unusually concrete for an influence story: named on-the-record sources (Brad Smith, Maggie Johnson, Randi Weingarten, Lois Weiner, Adriane Fugh-Berman, Bryan Twarek), hard figures ($23m, 1.8 million members, 59% device share, $170m), and a stated reporting base of 100+ interviews and hundreds of documents. It is capped by provenance: everything in this cluster is one aggregator's retelling of a New York Times excerpt, the underlying documents are not reproduced, and the load-bearing pharma comparison is an analogy rather than a measured finding.
Channel is live at national scale, with visible rollbacks
Adoption is directly observable rather than projected: a 1.8-million-member union running vendor-funded AI training, a 200-teacher Copilot session, vendor-authored AP courses in circulation, Chromebooks at 59% of US school device shipments, and millions of children through Google's privacy lessons. Score is held below the top band because counter-adoption is documented in the same material — Los Angeles Unified halting device issuance for prekindergarten through first grade, Weingarten calling for elementary schools to pause AI use, and the article's statement that a decade of classroom technology spend has generally not improved outcomes.
Mildly overstated: mechanism documented, causal reach asserted
The concrete elements — dollar figures, market share, named quotes, district rollbacks — are proportionate to their evidence, and the piece deliberately undercuts vendor benefit claims by noting outcomes have generally not improved. The overstatement is in reach rather than fact: the pharma-capture analogy and the 'nearly every step of the supply chain' conclusion do causal work that the presented material supports by illustration (no Raytheon physics course, near-identical handbook lines) rather than by measurement, and per-member funding of roughly $13 is a thin lever for the influence attributed to it. Small positive gap.
Dense and disclosed on nearly every side
Financial interest is explicit almost everywhere: three model vendors funding the union programme they will be trained on, Apple and Microsoft authoring AP courses in their own language and game platform, Code.org funded by Microsoft and Google while lobbying states, and the Computer Science Teachers Association's biggest donors including Amazon, Google and Microsoft even as its vice president objects to industry-driven curricula. Vendor executives themselves frame their role as philanthropy filling public failure. Countervailing incentives are also present and disclosed — the reporter's article is an adapted excerpt promoting her own book, and this cluster's publisher is aggregating another outlet's investigation for traffic.
Moderate: strong underlying reporting, single-publisher relay
Confidence is limited chiefly by cluster structure. There is exactly one source item, from an aggregator summarising a New York Times excerpt; no second publisher, no primary document, no vendor or union statement obtained independently. Internal consistency is good and the figures are checkable arithmetic, and the named on-the-record quotes plus concrete adoption facts keep confidence above the midpoint, but any error in the relay would propagate unchecked.
invest
Apple clears two bad bets off the books a week before Ternus takes over1 distinct publisher
leadership
A $110B demand side: the first deduplicated count of what AI buyers actually pay1 distinct publisher
invest
A $2B bet that voice replaces the text box, not just the keyboard3 distinct publishers
invest
The AI moat is now a balance sheet, so price the financing and not the model1 distinct publisher
Distinct publishers with included, body-backed reporting in this cluster.
1 article · August 25, 2026