They Saw the Speed. They Forgot to Price the System.

I. The Promise Arrived First

Every executive heard some version of the same pitch this year: AI would move faster than people, cost less than payroll, and run without complaint at three in the morning. Boards approved budgets on that promise. Ninety-seven percent of executives say their company deployed AI agents in the past year, and CEOs have taken direct ownership of the strategy — seventy-two percent now call themselves the primary decision-maker on AI, double the share who said so a year earlier. The pitch was speed. Nobody priced the system that speed would require.

II. The System Was Not Priced

The bill arrived faster than anyone modeled. Uber burned through its entire 2026 AI coding budget in four months, with roughly seventy percent of its committed code now originating from AI tools — and the company’s own COO has publicly conceded that usage hasn’t yet shown a clear link to shipped value. Microsoft, which writes up to thirty percent of its own code with generative AI, told engineers in one major division to stop using an AI coding assistant because the bills had become unsustainable. One unnamed company reportedly ran up a $500 million usage bill in a single month after management forgot to set a spending cap. None of this is a story about AI failing. It’s a story about companies pricing the tool and skipping the infrastructure, governance, and oversight required to run it responsibly at scale.

The prices themselves aren’t even real yet. Leading AI labs are pricing usage below what it costs to serve, subsidizing access with venture capital to buy market share while the true cost of the infrastructure is still being worked out. That subsidy is starting to unwind. When it fully does, the gap between what companies were told AI would cost and what it actually costs will widen again — and most budgets built this year were never built to absorb that.

The human cost inside these companies is its own line item, rarely put on a spreadsheet. More than 115,000 tech workers have been laid off in 2026 across over 150 companies, with the stated rationale almost always framed as efficiency and reallocation toward AI. At the same time, nearly three-quarters of CEOs report real stress or anxiety about their own AI strategy, and a majority privately admit that strategy is more performance than substance. A tool that makes leadership anxious while it eliminates jobs is not, by definition, a cheaper system. It’s simply a system whose true cost hasn’t been fully invoiced yet.

III. The Copyright Bill Entered the Room

This week, The New York Times and the Daily News asked a federal judge to sanction OpenAI, alleging in a court filing that the company withheld evidence relevant to a landmark copyright infringement trial over how its systems were trained on millions of news articles. The Times has reportedly already spent more than $28 million fighting AI companies in court — a legal budget most newsrooms could never absorb, spent defending a business model the newsroom didn’t choose to compete against. This isn’t a story about whether AI can summarize the news. It’s a story about who absorbs the legal cost when a company builds a product on material it never licensed — and whether “we didn’t ask” is a business model or a liability still awaiting its court date.

The industry’s own settlements are the clearest price tag available. Anthropic agreed to pay book authors $1.5 billion for training its models on pirated work — according to court filings, a landmark figure, though still a small fraction of the company’s overall valuation. Whatever a company saves in payroll by automating a task, it may eventually spend multiples of that figure defending how the underlying system was built. That’s not a hypothetical. It’s a line item companies are only now learning to forecast, and the forecasting gets more expensive every quarter they wait.

IV. Creators Are Building Gates

Creators didn’t wait for the courts. Cloudflare announced new defaults that block AI training and agent crawlers on ad-monetized pages unless site owners explicitly allow them, rolling out across the wider internet by mid-September. Patreon is the anchor partner, blocking AI training crawlers at the network level across its entire platform. Patreon’s founder was blunt about the terms: creators deserve credit, compensation, and consent, and platforms that won’t offer all three don’t get access to the work at all.

This is the same pattern KMOB1003 has tracked across music licensing, AI-cast performers, and AI-generated likenesses this year: the industry doesn’t wait for permission, so the people whose work fuels it are building the permission structure themselves, one platform at a time. Cloudflare’s own data shows why the urgency is real — for every referral an AI crawler sends back to a publisher’s site, it may crawl that site tens of thousands of times without ever sending a reader in return. That ratio is the actual cost of “free” training data, paid entirely by the people who made it, long before any court decides who owes what.

What makes this moment different from prior rounds of tech disruption is the speed of the countermeasure. Robots.txt was a polite request that most AI crawlers simply ignored while courts moved slowly. Blocking at the infrastructure layer — where roughly a fifth of the internet’s traffic already passes through a single provider — is a different kind of leverage entirely. It doesn’t wait for a verdict. It changes the terms of access today, for everyone building on top of that infrastructure tomorrow.

V. Control Was Always More Expensive Than Labor

Here is the KMOB1003 read on all three signals together: capital was never only chasing cheap labor. It was chasing controllable systems — and control turns out to be the most expensive part of the entire transaction. A worker negotiates a salary once a year. A compute bill renegotiates itself every time usage spikes. A worker’s consent is a single conversation. A creator economy’s consent, once you multiply it across millions of authors, photographers, musicians, and publishers, becomes a legal and technical infrastructure problem that companies are still building court cases and crawler blockers to solve. Speed was cheap. Control was not. Most leaders bought the first and are only now discovering the invoice for the second.

VI. Ask the Better Question

None of this is a reason to abandon the tools. It’s a reason to change the order of operations. The operators who come out ahead in this cycle won’t be the ones who adopted AI fastest. They’ll be the ones who asked, before the pilot ever left the sandbox, what the tool would make them responsible for — legally, financially, and to the people whose work made the tool possible in the first place. The question was never only what AI can do. The question was always who pays for what it changes. Ask that one first, and the invoice stops being a surprise.

The Operator’s Bookshelf

KMOB1003 READS

These belong beside this article because they ask the same question in different rooms: who gets paid, who holds power, and who decides when a digital system absorbs human contribution.

Who Owns the Future?

Jaron Lanier

Lanier helps explain why the internet’s “free” material was never free — it was human value waiting to be accounted for. Long before this year’s AI-cost reckoning, he was already asking who gets paid when a system learns from millions of uncredited contributions. It’s the same question sitting underneath every headline in this piece, just asked a decade earlier.

See Why We Recommend It →

Power and Progress

Daron Acemoglu & Simon Johnson

This book belongs beside any serious AI conversation because it refuses the fantasy that technology alone decides the future. Acemoglu and Johnson make the case, with a thousand years of evidence, that who benefits from a new technology is a choice made by institutions and power, not an automatic outcome of the tool itself. That argument is exactly what’s being tested in real time by every compute bill, lawsuit, and crawler block in this piece.

See Why We Recommend It →

KMOB1003 may earn a commission from qualifying Amazon purchases.

Disclosure: KMOB1003 may earn a commission from qualifying purchases through select partner links. Editorial coverage is produced independently.