# The Curriculum You’re Eliminating Divya Jain is right. Jain is a software engineer at Google and one of the sharper technology writers working today. In her February 2026 essay “The Myth of the Surplus Employee in the AI Era,” she argues that companies cutting headcount in the name of AI efficiency are committing strategic suicide. When engineers are freed from boilerplate work, they tackle architectural debt and move from maintaining to disrupting. The value is not the paperwork. It is the human outcome. Judgment, she argues, is the ultimate human multiplier — and judgment is built through years of calibrated experience, through thousands of solved problems and navigated crises. I want to extend her argument to a problem she does not address: what happens when that experience is never acquired in the first place? Picture a third-year associate at a fourteen-attorney firm. Smart. Trained at a top school. Reliable. This year, she used an AI platform to draft a motion for summary judgment in half the time it used to take. The structure was clean. The citations verified. The argument flowed. Something was wrong with the argument. Not the citations — a hallucination check would catch those. The underlying reasoning didn’t hold under pressure. It was the kind of flaw a junior attorney who had spent two years drafting briefs from scratch would catch in her own work. This associate had spent those two years reviewing AI output. The partner approved it anyway. Seventeen other matters sat on her desk. This is not a story about AI hallucinations. It is the problem Jain identifies — judgment as the irreplaceable multiplier — applied to the profession that will need it most. ## The Data SignalFire analyzed hiring at major technology firms and maturing startups between 2019 and 2024. Entry-level hiring dropped 50 percent across engineering, sales, marketing, operations, and legal. The decline was consistent across every function they measured. The Burning Glass Institute tracked job postings requiring three years of experience or less. In software development, those postings fell from 43 percent of the market to 28 percent between 2018 and 2024. In data analysis, from 35 percent to 22 percent. In consulting, from 41 percent to 26 percent. Total postings stayed flat or increased. Senior hiring remained stable. Companies are not hiring fewer people. They are skipping new graduates entirely. Revelio Labs found that entry-level postings plunged 35 percent between January 2023 and June 2025 — more than 100,000 fewer monthly job postings — with AI identified as a key contributor. None of these datasets break out legal separately. They don’t have to. The pattern is identical across every knowledge-work sector AI has penetrated. The legal profession is not exempt from the math. ## The Curriculum Was Never the Output Matt Beane has spent over a decade studying how workers develop skill. He’s an assistant professor of technology management at UC Santa Barbara, a digital fellow at Stanford and MIT, and the author of The Skill Code: How to Save Human Ability in an Age of Intelligent Machines. His central finding: expertise requires three conditions. Challenge — working near the edge of current capability. Complexity — engaging with full systems rather than isolated tasks. Connection — learning alongside someone who knows more. He calls this the skill code. His published summary of the current state: “We have started a war between technological productivity and human skill, and skill is losing.” Researchers who study expertise distinguish between explicit knowledge — what can be written down — and tacit knowledge — what can only be absorbed through practice. You can read every brief filed in the Seventh Circuit and still fumble your first deposition. You can study motion practice frameworks and still misjudge when a summary judgment actually has a chance. Document review taught junior associates what relevance looks like. Drafting interrogatories taught them how parties resist disclosure. Building research memos taught them the difference between a case that says what you need and a case that merely sounds like it does. When AI handles these tasks, the output may be faster. The learning disappears. ## The Seniority Cliff One researcher frames this as the risk of a “seniority cliff” arriving in five to ten years. Seniority is not a function of age. It is the accumulation of experience that produces reliable judgment under pressure. If associates never grapple with foundational work because AI handles it automatically, they never build the intuition required for senior roles. Firms that stop building that foundation today are eliminating their senior attorneys of 2035. Consider what that produces in practice. Courts are already documenting attorneys who cannot evaluate AI output — who submitted hallucinated citations because they had no independent basis to recognize the work as wrong. By December 2025, researcher Damien Charlotin’s database had reached 979 documented judicial decisions involving AI-generated errors. The common thread across every case: the attorney used the tool, did not understand its limitations, and submitted the output without the judgment to catch the flaw. My prior piece “When Attorneys Stop Checking AI Work” examined that sanctions record in detail. Comment 8 to Model Rule 1.1 makes technological competence an ethical obligation. What that standard requires is not just knowing how to run a prompt. It requires the judgment to recognize when AI output is analytically wrong — when the reasoning sounds plausible but won’t survive a motion hearing. That judgment is built through years of foundational work. It cannot be imported from a training module. Seniors catch errors that juniors miss. That is exactly what judgment looks like. But today’s seniors became seniors by doing the work that today’s AI now handles. Remove the foundational work from tomorrow’s seniors’ early years, and no one is left to catch the errors in 2035. ## The Optimist Case, Stated Fairly Proponents of AI workforce benefits are not wrong. Their argument is incomplete. PwC’s 2025 Global AI Jobs Barometer found that AI makes workers more valuable, not less. Wages are rising for automatable jobs. Stanford researchers found that skills in prioritizing, teaching, and communication will grow in value as AI handles more analytical work. The technology reshapes roles toward judgment and creativity. All of this is accurate. All of it is contingent on workers already having the judgment to do complex work. The strongest version of the optimist argument: technological transitions have always displaced some jobs and created others. The calculator made mathematicians more productive, not obsolete. Word processors eliminated typing pools but created desktop publishing. The pattern holds over centuries, and the legal profession has absorbed every prior technological transition without losing its pipeline. That pattern holds under one condition: foundational skills remain transmissible. Mathematicians who used calculators to solve harder problems learned arithmetic first. The technology accelerated their work. It did not replace their foundation. If calculators had arrived and schools had stopped teaching arithmetic because the machine handles it now, there would be no mathematicians a generation later. There would be people who can operate calculators but cannot recognize when the output is wrong. That is the trajectory law firms are building toward if they celebrate AI efficiency without deliberately maintaining the conditions that produce judgment. ## The Diamond Age Problem In Neal Stephenson’s novel The Diamond Age, a device called the Young Lady’s Illustrated Primer educates a child through interactive scenarios that adapt to her development. The Primer does not give answers. It creates situations requiring judgment, escalates complexity as the learner grows, and never resolves problems the learner should resolve herself. Its purpose: to raise someone capable of thinking independently. Most firms are not using AI as the Primer. They are using it to skip the primer entirely. Medical education is ahead of the legal profession here. Researchers have developed AI tutors that guide learners through clinical cases with real-time feedback on reasoning, creating deliberate practice environments that build diagnostic judgment. The model is not to hand the resident an AI-generated diagnosis. It is to use AI to create more complex training scenarios faster. The learning is the output, not the answer. Law firms have the infrastructure to build something equivalent. Most are allocating it toward efficiency metrics instead. ## Three Things You Can Do Before Friday The argument that AI efficiency is consuming seed corn — made by Divya Jain, a Google software engineer and widely read technology writer, in a February 2026 essay and extended here — applies with particular force in legal practice because the work AI replaces in a law firm is not merely productive. It is pedagogical. Document review is not a task. It is where associates learn what matters. Three adjustments that do not require a strategy retreat. First: when AI drafts a research memo for an associate, require the associate to produce an independent analysis before reviewing the AI output. Not to catch errors — to build the habit of forming judgment first. The AI version becomes the check, not the starting point. An associate who has never wrestled with the question independently has no basis to evaluate whether the AI answer is right. Second: designate one foundational task per associate per matter cycle that must be completed without AI assistance. Document review. A research question. A motion section. The output is not the point. The exposure is. Third: revise your associate evaluation criteria to include AI supervision, not just AI use. An associate who can use AI efficiently is valuable. An associate who can identify when AI output is analytically wrong — when the reasoning sounds plausible but fails under examination — is worth considerably more. “AI Won’t Take Your Job. The Attorney Who Uses It Better Will.” made this point for senior attorneys. The same logic applies to building the pipeline that produces them. These are not arguments against AI. They are arguments for deliberate investment in the judgment that makes AI supervision possible. ## The Seed Corn There is an old saying in agriculture: you can eat your seed corn, but only once. Seed corn is the portion of the harvest set aside for next year’s planting. Eating it solves this year’s hunger. It also guarantees there is no crop next year. Foundational legal work — document review, research memos, first-draft motions — is seed corn. It is where tomorrow’s judgment develops. Eliminating it is efficient. It is also irreversible if you wait long enough to notice the cost. Model Rule 1.1 requires competence. Competence in 2035 will require attorneys who can supervise AI output under pressure — who can distinguish a plausible brief from a sound one, a fabricated authority from a binding one, an argument that reads well from one that holds. That kind of attorney is built through years of work that AI can now handle faster. The question is not whether to adopt AI. The question is whether to also solve the problem AI creates. — This blog provides general information for educational purposes only and does not constitute legal advice. Consult qualified counsel for advice on specific situations. ## About the Author JD Morris is Co-Founder and COO of LexAxiom. With over 20 years of enterprise technology experience and credentials including an MLS from Texas A&M, MEng from George Washington University, and dual MBAs from Columbia Business School and Berkeley Haas, JD focuses on the intersection of legal technology, cybersecurity, and professional responsibility. Connect: LinkedIn: http://www.linkedin.com/in/jdavidmorris | X: @JDMorris_LTech | Bluesky: @JDMorris-ltech.bsky.social ## References SignalFire. “State of Tech Talent Report 2025.” SignalFire.com. Analyzed hiring at 15 largest tech companies and maturing VC-backed startups, 2019–2024. 50% decline in new-graduate hiring confirmed across all core business functions. Burning Glass Institute. “No Country for Young Grads.” July 2025. Entry-level postings in software development, data analysis, and consulting, 2018–2024. Revelio Labs / Bloomberg. “Is AI Responsible for the Rise in Entry-Level Unemployment?” August 4, 2025. 35% decline in entry-level postings, January 2023–June 2025. Beane, Matt. The Skill Code: How to Save Human Ability in an Age of Intelligent Machines. Harper Business, June 2024. Three-C framework (challenge, complexity, connection). “We have started a war between technological productivity and human skill, and skill is losing.” MIT Sloan, 2024. PwC. 2025 Global AI Jobs Barometer. Productivity and wage effects of AI adoption. Charlotin, Damien. AI Hallucinations in Judicial Decisions Database. 979 documented decisions as of December 2025. ABA Model Rules of Professional Conduct, Rule 1.1, Comment 8 (Technology Competence). Stephenson, Neal. The Diamond Age: Or, A Young Lady’s Illustrated Primer. Bantam Spectra, 1995. Jain, Divya. “The Myth of the Surplus Employee in the AI Era.” Medium, February 2, 2026. Jain is a software engineer at Google. Prior Blog: “AI Won’t Take Your Job. The Attorney Who Uses It Better Will.” Parts 1–2 (Morris Legal Technology Blog). AI competence under Model Rule 1.1. Prior Blog: “When Attorneys Stop Checking AI Work” (Morris Legal Technology Blog). Sanctions record and hallucination database through 2026. Prior Blog: “Every Failed AI Project Breaks the Same Rule” (Morris Legal Technology Blog). Gall’s Law and Johnson v. Dunn AI governance failure.

Originally published on LinkedIn Newsletter: The Technology Blind Spot