Codex of the Lesser Machine-Spirits

The SLM Manifesto

Verse V · Backlash

Backlash

Evidence made the technical case and Costs made the economic one. This chapter makes the human case, and it does not require a single argument of my own, because the public has already made it, in polling, in layoff data, and in courtrooms. The frontier promised a substitute for skilled judgment. Checked against real coverage rather than press releases, the substitute has been failing publicly, getting walked back quietly, and, in at least one field I know from the inside, drawing sanctions from judges who caught the difference.

What the public actually thinks

Pew Research surveyed 5,119 US adults in 2026 and found roughly half now use an AI chatbot, up from about a third in 2024.[1] Adoption climbed. Trust did not follow it. Forty percent of the same respondents expect AI to make society worse over the next twenty years, and 67 percent have little to no confidence in the US government's ability to regulate it, up from 62 percent just two years earlier.[1] Other 2026 polling points the same direction: Quinnipiac found 76 percent of Americans trust AI-generated information hardly ever or only some of the time, and an Economist and YouGov poll found more than 70 percent of Americans, across party lines, think AI is advancing too quickly. People are using this technology more and believing it less, at the same time. A trusted replacement for human judgment does not usually look like that.

Pew Research, 2026. Adoption is rising. Trust is not.

The layoffs that were never really about AI

AI-attributed layoffs went from roughly 18,000 in 2024 to more than 100,000 in 2025, and the first half of 2026 alone topped 150,000, about half again as many as all of 2025 combined. As of mid-2026, 56 percent of tracked layoff events, 150 of 267, name AI or automation as a cause.[2] The same year, Alphabet, Microsoft, Meta, and Amazon are on track to spend nearly 700 billion dollars combined on AI infrastructure while cutting tens of thousands of jobs, Meta alone cutting roughly 8,000 roles in May with an internal memo tying the cut directly to offsetting its own AI capital spending.[2] If AI were actually doing the work these layoffs credit it with, that spending and those cuts would move together. They do not survive a closer look. A Gartner survey found companies cutting jobs over automation regardless of whether the technology was generating any return, and cuts happened at roughly equal rates whether the AI in question was working or not.[2] An NBER working paper found 90 percent of executives report AI has had zero employment impact at their own firm, even as their own companies announce AI-attributed cuts.[2] Cognizant's own chief AI officer put it plainly: sometimes AI becomes the scapegoat from a financial perspective. Sam Altman used a blunter term for the same pattern: AI washing.[2] AI did not start replacing workers in 2026. Executives just traded one cover story for another, swapping the old word, restructuring, for a newer one that plays better on an earnings call.

AI-attributed layoffs, thousands of workers. 2026 figure is the first six months alone.

Ford

Ford is the clearest case of what happens when the substitution actually gets tried instead of just announced. The company had cut 5,300 salaried positions since its 2020 employment peak, part of 20,000 white-collar jobs eliminated across Detroit's three automakers over the same stretch, and had replaced experienced quality-inspection engineers with an AI system meant to do the job for less. The experienced engineers left before their expertise could be encoded into the systems meant to replace them, and the AI, trained on nothing of real substance, produced an inferior product that cost Ford billions in warranty and recall expense.[3] Ford's VP of vehicle hardware engineering, Charles Poon, put the mistake in his own words: the company had assumed that introducing AI and adjusting design requirements would, on its own, produce quality. It did not. Ford rehired, newly hired, or promoted 350 veteran engineers, the ones staff had started calling the gray beards, built a 40-person software QA team, and added more than 100,000 AI-powered automated tests, this time with the humans who actually knew what failure looked like rebuilding the data pipelines and reprogramming the systems they had originally been hired to replace.[3] COO Kumar Galhotra described their value directly: they hunt for failure points before a part ever reaches the plant floor. The result was not a wash. Ford topped JD Power's initial quality study for the first time in sixteen years, and CEO Jim Farley credited the rehiring effort with hundreds of millions of dollars in savings on warranty and recall costs alone.[3] Trying AI was not Ford's mistake. Trying to remove the judgment AI still cannot supply was, and fixing it meant paying experienced humans to come back and supply that judgment themselves.

Where the substitution gets tested hardest: law

No field makes the limits of current AI more measurable than law, because law has a built-in mechanism the rest of the economy lacks: a judge, on the record, checking the work. Lawyer Damien Charlotin's public database now tracks more than 1,300 cases globally where a court has formally flagged AI-generated hallucinations in a filing, and the sanctions keep escalating rather than tapering off as the tools mature.[4] In May 2026, the Alabama Supreme Court dismissed an appeal and barred an attorney from filing further briefs without co-counsel approval, calling his conduct egregious, after he cited a fabricated precedent, was warned, and then cited more nonexistent cases in the very next filing.[4] The same month, a federal judge in Oregon imposed the largest AI hallucination penalty in US legal history, 110,000 dollars, after lawyers submitted 23 fabricated citations and eight invented quotations.[4] In March 2026, the Sixth Circuit sanctioned two attorneys in Whiting v. City of Athens for more than two dozen fake citations across three consolidated appeals, ordering them to reimburse the other side's legal fees, pay double costs, the stiffest penalty available under the relevant appellate rule, and 15,000 dollars each in punitive sanctions.[5] A Manhattan court went further still in a separate case, ruling that a defendant who used a general-purpose chatbot to help build his own defense had waived attorney-client privilege over that strategy entirely. Every one of these is a licensed attorney, trained for years, using a tool marketed as ready for professional legal work, and getting personally sanctioned by a judge for trusting it. A lawyer who hands off a citation without checking it by hand is not looking at a technology on the verge of replacing them. The sanctions orders keep saying otherwise.

I am not writing that from the outside. I hold a Juris Doctor from the University of Miami School of Law and have been a member of the Florida Bar since September 2016, and I have spent the last several years as Lead Development Manager and then Acting Chief Technology Officer at Lexmata, building the legal-technology platform attorneys and their clients actually use. I have sat on both sides of this problem at once, as counsel and as the engineer responsible for the AI tooling itself, and both roles teach the same lesson. Current models earn their keep on a first draft, a structured summary, a starting point for research a competent attorney still has to verify line by line. Skip that verification and every sanctioned filing above is what waits on the other side. A field with this much formal, adversarial, judge-supervised checking is the clearest place to watch current AI's actual stage of development: useful help for someone already competent, no substitute for the competence itself, and the courts keep proving the difference one sanctions order at a time.

None of this argues that AI is useless. It argues against the specific claim this manifesto has opposed from its first page: that the technology is mature enough, right now, to justify replacing skilled people wholesale, on the strength of a press release rather than a verified result. The public does not believe that claim. The layoff numbers built to support it do not survive scrutiny. Ford tried it in hardware and had to pay experienced engineers to come back. Law tries it every week and produces a sanctions order. The judgment competent people supply is still the thing missing, and no amount of scale has supplied it yet.

Sources

  1. Pew Research Center. "Americans and AI 2026: Chatbots, Smart Devices and Views on Impact." June 2026. pewresearch.org. Chatbot adoption, expected societal impact, and confidence in AI regulation, from a survey of 5,119 US adults.
  2. The Interview Guys. "56% of 2026 Layoffs Now Blame AI, But the Companies Cutting Jobs Are the Same Ones Spending Billions on It." 2026. theinterviewguys.com. The layoff-attribution data, the hyperscaler spending-versus-cutting figures, and the Gartner, NBER, Cognizant, and Altman findings cited above.
  3. Toscano, J. "Ford Hiring 350 Engineers After AI Failed Shows Human Value In AI Era." Forbes, June 2026. forbes.com. The Ford quality-engineering failure and rehiring, with the Poon, Galhotra, and Farley quotes above.
  4. Fortune. "Would you hire the lawyer who just got sanctioned for using AI?" May 2026. fortune.com. The Alabama and Oregon sanctions cases, and Damien Charlotin's 1,300-case hallucination database.
  5. Sixth Circuit Appellate Blog. "Sixth Circuit Sanctions Attorneys for Fake Citations, What Does This Mean for Use of AI?" March 2026. sixthcircuitappellateblog.com. Whiting v. City of Athens, the fake-citation count, and the sanctions imposed.