Underwriting Stagnation: Taxpayers Pay for Schools to Stand Still
California Teachers Can Now Veto AI
The Union Gets the Final Say
California’s largest school districts have done something unheard of in American management: they have given a labor union veto power over whether a promising technology may ever be used to reorganize work.
On January 22, 2026, the San Francisco Unified School District (SFUSD) signed a one-year contract with OpenAI for up to 12,000 ChatGPT accounts for teachers and staff.
On March 10, 2026, one month after settling a four-day strike, SFUSD approved an agreement with its teachers union and announced that it had accepted the union’s AI proposal. That proposal barred AI use “to eliminate, reduce, or replace work traditionally performed by bargaining unit members without prior written agreement of the Union,” and barred using productivity gains from AI “as a basis for staffing reductions or increased workload expectations.” Weeks after buying the AI tool, the district agreed it could not use it to change how the work gets done. The district has yet to publish the final signed language.[1]
A few months later, Los Angeles Unified School District (LAUSD) did the same thing, in language we can read. Its agreement with United Teachers Los Angeles, titled “Use of Technology to Support Instruction,” says: “GAI [generative AI] shall not be used to replace Bargaining Unit positions or members in doing work historically and generally performed by Bargaining Unit employees without the express written agreement of the Union.”
Even in the world of labor-favored public sector labor relations in California, this is over the top. As a former public sector manager, I was accustomed to giving the union notice before changing how work was done, and to negotiating over how the change would land on employees — schedules, workload, assignments. A manager who owes notice still gets to decide.
A district that must obtain written permission does not decide at all. These clauses set no deadline by which the union must answer, no standard governing a refusal, and no appeal. A teacher’s worry that a district in fiscal trouble will reach for AI to cut headcount is not irrational. But provisions are already in place requiring districts to negotiate workforce reductions. A veto is a different instrument.
Experimenting vs. Standing Still
Meanwhile, the rest of the economy is working through the hardest management question of the decade—but not in the way the teachers unions’ demands suggest.
Employers are overwhelmingly not firing people over AI. Gartner found that only about one percent of studied workforce reductions was directly attributable to AI productivity gains. The Census Bureau found that 66% of adopting firms use AI solely to augment tasks, and that AI-related employment decreases occurred in just 2% of firms. Boston Consulting Group projects that 50 to 55% of U.S. jobs will be reshaped within two to three years: the same role, with very different expectations about how the work gets done.
Yes, jobs are being reshaped. Employers are grappling with what their employees’ work should look like now that this robust new tool is available—not how many people they can cut. They are examining which tasks move, which roles change, what a job description should look like three years from now. Nobody knows how this will play out. That is precisely why organizations must preserve the ability to find out.
Every major technological advance begins with uncertainty. No one knew in advance how spreadsheets would change accounting or how smartphones would change field work. Organizations learned by experimenting—not by negotiating permission to experiment. These agreements go far beyond opting for caution. They prohibit experimentation unless the union agrees first.
Private schools face no such restriction. The best private schools will study what AI can do and adopt it where it improves learning. These new union contracts are the taxi medallion of 2013—a legal restriction that protects incumbents, raises the price, and holds only until the alternative arrives. While the world moves forward, we are inflating a government education productivity bubble. Taxi medallion owners bore their own losses. Public school districts socialize theirs, sending the bill to students and taxpayers. The students who pay first are those whose families cannot afford another school. The taxpayers pay next—including those who leave and continue paying for the district they left.
The Students Lose
The greatest cost of restricting AI is not administrative efficiency. It is what students lose. If AI turns out to be even a fraction of the tutor it promises to be, the children who gain most are the ones whose parents cannot afford one. Until recently, only well-off parents, or those willing to uncomfortably stretch their budgets, could afford tutors for their academically struggling children.
Just as the internet brought world class libraries to the masses of children who did not live close to a library, AI brings the possibility of a versatile tutor. A tutor that will ensure that the precocious child is not held back by the class average and the tired school syllabus and struggling students without parental means can get the academic help they need.
That a school district would resist such technology is unconscionable.
Who Pays for Standing Still
LAUSD enrolled 408,083 students in 2024-25, down from 746,831 in 2002. Nearly half its zoned elementary schools are half-full or less. Per-pupil spending has nearly doubled in a decade, to roughly $27,000. About ninety cents of every dollar the district spends goes to people. The district is spending about $2 billion more this year than it takes in and has hired Ernst & Young to model school closures.
An organization can control costs in only two ways: by changing what it does or by changing how it does it. The freeze on AI effectively neutralizes the second lever in the context of an organization that is much more likely to expand its scope than trim it.
On June 16, 2026, LAUSD signed that lever away.
On July 2, 2026, the Los Angeles County Office of Education issued the district its first-ever Lack of Going Concern determination.
The teachers unions are quick to frame any critique of the status quo government educational system as an attack on teachers and an unwillingness to pay them what they deserve.
Sadly, it is the teachers union’s own bargaining agenda that restricts us from justly compensating teachers. State law puts every teacher on a salary schedule keyed to years of training and years of experience, and a district may depart from it only with the union’s agreement. So the teacher who excels is paid on the same scale as the one who does not, and the district cannot change that without the union’s permission. Now the district will need the permission of the same union before AI can do work its employees do today.
The district does not decide.
In a productive enterprise, technology makes a worker more productive, and the worker earns more because the output is worth more. That is a trade, and it provides a reasonable basis for a raise. These agreements reverse it and then some. The compensation increase comes first—an average of 13.86% at LAUSD, with nothing below 8%—and the productivity improvement that would have justified it is contractually prohibited. The bill goes to a taxpayer who had no say and to a student whose school is on a consolidation list.
There is a structural reason this keeps happening. In the private sector, a firm that refuses to adapt loses customers to one that does. Employer and workforce therefore share an interest in becoming more productive, because productivity is what makes higher wages possible. A government school district faces no such discipline. Its revenue arrives as per-pupil funding from the state. The people who bear the cost of stagnation are students and taxpayers, and they are not at the bargaining table. The organization across the table is an active participant in the elections that decide who sits or stays on the board.
Meanwhile, the California Teachers Association is working to spread the AI restrictions adopted in San Francisco and Los Angeles across the state. Its “We Can’t Wait” campaign aligns contract expirations in 32 districts covering 80,000 educators, and UTLA’s own platform lists a “ban on subcontracting and replacement of staff with AI or other tech.”[2]
Approving the Unknown in a Public Meeting
How does something so egregious as bargaining away one of the two levers a district has for controlling cost get approved in a public forum?
Before a school board may vote on a labor agreement, state law requires it to disclose in public “the major provisions of the agreement, including, but not limited to, the costs that would be incurred.” The superintendent and the district’s chief business official then certify that the district can pay for it.
The problem: Only the employee compensation provisions are quantified and identified. The fact of surrendering authority critical to the financial stability of the organization is not considered. With the state of our knowledge today, nobody could responsibly quantify the savings AI might produce for a school district.
But a district can recommend that its board forbid finding out.
Three questions belong in front of every governing board before it ratifies a labor agreement: what operational authority are we giving up, how much is it costing us to give it up, and what would it cost to buy it back?
History will not record that AI failed California’s schools.
It will record that California negotiated away the right to find out.
Mark Moses is a senior fellow with the California Policy Center. He has thirty years of experience in local government administration and finance. His book, The Municipal Financial Crisis – A Framework for Understanding and Fixing Government Budgeting, was published by Palgrave Macmillan and is available from major online booksellers.
https://munifinanceguy.com/ · X/Twitter: @MuniFinanceGuy
[1] The account here rests on the union’s proposal text, the district’s own update, and press reporting.
[2] These are proposed contract terms, not statutes — but the law is moving the same direction, and it reaches further. Since January 1, 2026, AB 339 has required a local agency to give its unions 45 days’ written notice — with anticipated cost and a draft solicitation — before issuing a request for proposals or even renewing an existing service contract. That one applies to cities, counties, and special districts; school districts and the state are excluded. AB 2656 would extend the same 45-day notice to developing, purchasing, implementing, using, or requiring the use of generative AI within the scope of a represented job classification — and it carries no such exclusion. It reaches every public employer in the state: state agencies, cities, counties, special districts, K-12 districts, community colleges, the University of California and CSU, the trial courts, and transit districts. It passed the Assembly in May and sits in Senate Appropriations; the session ends August 31. SB 951 would require 60 days’ notice before any “technological displacement,” give affected workers a right of first bid on other positions, and bar discharge without “reasonable and substantiated cause” during the notice period. It passed the Senate 28-9 and is in Assembly Appropriations.
Notice is not consent, and neither bill grants a veto. But notice is the on-ramp. Once the law establishes that deploying a tool is a labor event requiring advance notification, the demand at the next bargaining table is not for notice. It is for the answer.