AI and the Weaponization of Bureaucracy

It took 8 hours in 2022, 30 minutes in 2026. It was the same appeal for the same house in the same county. The resulting document was better the second time, and I didn’t write it. AI did.

Weaponized Bureaucracy as Business Model

We have all heard of or experienced the infuriating maze of health insurance. Vehicle insurance is similar. Endless requirements, buried in fine print; masses of paperwork and forms to fill out; manual effort to track down some specialized version of a document. I won’t say that these systems were designed adversarially, but they certainly feel adversarial in their current form.

Though I will talk about it here as a business model, bureaucracy is also weaponized in government. Indeed, just this week, Ezra Klein interviewed Annie Lowrey on just this topic, highlighting that liberals and conservatives alike have engineered systems that put major burdens on the very people the systems are supposed to benefit, and that these control systems fall disproportionately on lower-income or otherwise time-burdened people:

I think that they might not conceive of it that way because we are not beating somebody over the head with: Hey, you got a home mortgage interest deduction, for which you didn’t have to pee in a cup.1

And this is not a new phenomenon. Lowrey first coined the term “time tax” in a 2021 Atlantic article2 and has gone much deeper in her book of the same name3. This is far from just a business model question. It is a question of how we, with good intent or ill, construct systems with a goal of leading people into dead ends or, at best, lead them into such a labyrinth of specific requirements that only the most dedicated and fluent-in-government-speak can successfully navigate the system.

This effect shows up everywhere. And it was parodied in the Rebate Gas episode of Nathan for You, where they assumed by creating a sufficiently onerous rebate submission process, they could entirely prevent any redemptions:

Property Value Assessments

I live in Oakland, CA, so our property taxes are subject to Prop 13, which deserves a write-up all its own. In short, Prop 13 divides Oakland into two:

  • Long-held properties that are assessed on a base year (often decades in the past) with a cap of 2% value growth in each year since, meaning that these properties are often valued well below their actual value for tax purposes.
  • Recent purchases whose base year is recent and is paying close to or at market rate

Two identical houses on the same block pay property taxes that differ by 5x, determined only by the year of their last transaction. Prop 13 is a huge driver of inequity in this city, and it is encoded in state law. (as an example, one random property I found in Oakland has a tax value of $269,158, while Redfin estimates its value at $1.5 million, meaning the owners are paying 82% less tax than the new owner would be if they sold it today)

Every year, we get a property tax assessment letter, and every year it is absurdly higher than any local comparison. This makes sense, because the government has an incentive to get as much property tax from its citizens as possible. But we have recourse as taxpayers. Alameda County has an “informal” appeal process and a formal one under Prop 8. The informal process requires finding comparisons within a fairly strict set of rules (sold within a narrow window, very similar in terms of spec to my house, etc.). Several years ago, I appealed for the first time on principle. Of course, the informal appeal was rejected after I spent hours on Zillow and Redfin and Google finding comps and making arguments. Then I paid the $50 fee to file the formal appeal to go through the process again a second time with an even more esoteric set of requirements, only to hear nothing for 3-6 months. I finally received an offer of settlement from that county, which I accepted, even though I assume that, had I pursued it to the end, I would have gotten more back 2 years later. I had had enough. Now I am in the process again, submitting yet another informal review. I’ll talk about this in a later section, but spoiler: it is not taking hours anymore.

Vehicle Insurance

Similar to health insurance and taxes, auto insurance is notoriously paperwork-laden. I have been in 2 accidents in the last 10 years. Both were deemed 0% my fault. In both cases, I had to go back and forth dozens of times with the adjuster to determine the value of my vehicle (both were totaled) providing documentation, comps, and a million other little details. In my 2018 accident, it was a manual slog to respond to each of these, taking hours and ultimately leading to my throwing up my hands and accepting the then-current offer. In 2026? Each response took me about 3 minutes of actual work. I had Claude scan comps for my car, give me exact phrasing to use with insurance, and write the emails to the adjuster. It took a painful process and made it simple.

Health Insurance

Looking at healthcare, there is clearly an all-out war between caregivers and insurance companies, with practices completing 40 prior authorizations per week and 32% of physicians reporting that prior authorizations are often or always denied 4. This would make sense if these are nonsense claims that should have been denied, but they are not. Upon appeal, 81.7% of these initial denials are overturned, highlighting that a real evaluation never took place before the immediate initial rejection5. No system should have an 82% false negative rate, especially one that is supposed to protect people from financial ruin when they have health concerns.

These insurance companies, hospital systems, and doctors hide behind each other’s paperwork. It seems like every time I go to the doctor for more than a checkup, I enter a billing nightmare that no one seems to be able to understand, and I end up footing the bill. Insurance companies with Explanations of Benefits that require going 6 documents deep to decode them. Doctors who break bills into “facility,” “doctor,” “other doctor” for anything complicated. I thought I was going in for a single procedure, why am I getting billed separately for anesthesia from a totally different organization? AI can help decode all of this and prevent the buck-passing that seems to define the healthcare industry today, and it can do it fast.

AI as Antidote

But that might be changing. It turns out that AI is extremely good at filling out forms and pulling together disparate data. This year, Claude was able to find 3 properties, compare them, and write a response letter with little input from me, and I was able to refine it in just a few minutes to be exactly what I wanted. When that is inevitably rejected, I will do the same thing for the formal appeal. This is great for me, but it is very bad for the county. They rely on the volumes of these appeals being small, whether intentional or not, waning as someone gets deeper into the process and becomes more and more resigned to the status quo.

You know what doesn’t get tired? AI. In all three of these domains, and more to come I’m sure, I can generate collateral in incredible volumes. With some clever sourcing, I can get all of the data I need to make my case. I can fill out forms in minutes. I can write letter after letter, and the county or company has to respond to them. While I am sure they will start to use AI, too, the balance of power is still in my hands because I represent one real-estate parcel of 468,000 in the county6. It is a certainty that the county cannot possibly respond to all of these manually. More important, though, is that the county’s finances would suffer immensely if all of these appeals went through.

The Oakland Math

We need to do a little math to see what the effect might be. About 40% of Oakland properties are functionally capped by Prop 13, so we can assume that those won’t be re-appraised under Prop 8 7. Only recent purchasers realistically have a Prop 8 claim, and only those who bought before prices fell. This is a minority of properties. But under Prop 13, it is a huge percent of the total property tax base because the rest of the roll is taxes on 1990s property values. So there is a tension in the analysis below: excluding long-held owners raises the financial exposure while narrowing to recent purchases at market peaks reduces it. I can’t be sure of the net without the assessor’s full roll composition, so the figure below is illustrative.

Properties that can be reassessed are predominantly higher value properties, so let’s assume that 50% of all of the property value in Oakland is re-appraisable. In my case, my last appeal yielded a 15% reduction in our property value, a truly insane discrepancy in a single year. Perhaps obviously, we bought near the peak of the market, and a decline in property values in our area came immediately after. Under Prop 8, the county is required to reassess property values, and for every year I have lived in this county, I have had to make a specific request for a reduction in value. I file the informal appeal letter, which disappears into the void with either no response or a rejection (time without AI: 6-8 hours; with AI: 30 mins). Then I file the formal appeal and hear nothing for 3-6 months. I then get a letter back asking for additional comps with specific additional requirements on them (time without AI: 4-6 hours; with AI: <30 mins). Then I get a settlement offer to reduce my value but not by that much. Historically, I have accepted this offer because I want to be done with the process, but going forward, I may have a higher bar because generating evidence becomes so easy and inexpensive that it is in my interest to hold out for the maximum.

The effect on the county could be terrible, though, if this becomes commonplace. Given the numbers above, if such over-valuing is a systematic issue, then the same appraisal, if exercised by everyone, could lead to a 7.5% decrease in property tax revenue to Oakland. No resident of Oakland or Alameda County wants this. That revenue pays for our schools and roads and fire departments, but a tax roll that only reflects the true value of property for those who jump through hours of technical and procedural hoops is an unequal, unfair system if certain people lack the time and expertise to advocate for themselves. This is far worse than peeing into a cup; it is a shadow tax that is enforced only by the inaction of administrators, not by legislation.

But the fact remains: the county has no obligation to do me favors. It is obligated by the state constitution to reassess my home’s market value when it falls, but it does not do so unless I ask, and to them, asking costs money, time, and participation in an opaque system in which I have little-to-no recourse.

The House Makes the Rules (and always wins)

Does this mean that AI can solve the problem in the long run? If anyone was making the argument that these sorts of controls are necessary and that these forms and processes are just the price of doing that business, then AI might actually solve the problem. MCP connectors and a Claude/ChatGPT subscription and model templates would create a much simpler system for gathering this information. But even the most casual observer knows that these forms being necessary is nonsense. We have every ability without AI to make these processes smooth, and we choose not to. The powers that be, whether the tax man, or the benefits delivery organization, or any other institutional power, continue to throw up barriers because it is in their interest to do so. Insurance companies do not talk about care delivery or simplicity on earnings calls; government organizations don’t get graded on increasing outflows; county assessors don’t want to see property values fall. They will ultimately adopt AI to their ends as well8.

So will AI fix the problem? No, probably not in any durable way. But AI does change the game. Organizations, particularly large ones, struggle to adopt new practices. AI is the first time that the consumer has a tool that can quickly overcome at least some of the barriers that The House throws up. Will it be enough? Probably not to make durable change. Will it be enough to throw sand in the gears of existing systems, response commitments, and care decisions? It may well become enough for that. And in a market where the consumer seems to be the last beneficiary in a stack of extractive practices, any reset is a valuable thing.

Those Who Need It Most Get It Last, If At All

I am very hopeful about AI when I think about it in my own life, but I also realize that I am one of the people who has always been good at paperwork. It is a weird thing to say, but somehow I have learned to relish working within a system to get what I need, so AI is enhancing the things that I am already doing. Prop 13 sorted once on sale date. But the appeal process applies a second sorting: who has the time. I have been on the winning side of this second sort, and AI just reduced my hours required. If AI stays with me and my ilk, it becomes a third layer of sorting. Rather than unrolling the bureaucracy, it advantages me yet again while failing to help those who would benefit most.

Word of AI and what it can do is not getting out fast enough. Sure, panic and worry about jobs is getting out. Datacenter hatred: public. Copyright suits: underway. The public profile of AI outside of a few major AI centers is “super-Google” or “interesting novelty.” But it isn’t. It is a tool that can pay enormous dividends to lots of people when used the right way.

My biggest concern is that those who will benefit most have no idea that even a free ChatGPT or Claude subscription can transform their ability to respond to a wall of red tape. AI will certainly take jobs, or at least change them. AI will certainly consume energy, though that is for another post because the level of disinformation out there is enormous. But my concern is that AI is going to become yet another equality issue. The wealthy will use it to make their lives easier, automate things, and make better financial decisions. Those who are unaware or see it as “not for me” will never see its benefits.

The optimist in me has hope that these tools will only become more available and the default way to fight bureaucratic weaponization. The version of AI that helps the most people is not the one running on my laptop or on a Mac Mini in some coder’s garage. It is a legal aid clinic that can write briefings faster or dispense solid advice tailored to someone’s specific situation. It is a patient advocate that can feel more like a concierge doctor by decoding an EOB in seconds. It creates the potential for a world where the marginal cost of high quality advocacy and counsel falls precipitously. Whether it reaches those who need it is a question of distribution, not evangelism, and distribution does not happen on its own. It can happen, though, and I hope it will.

Footnotes

  1. https://www.nytimes.com/2026/08/25/opinion/ezra-klein-podcast-annie-lowrey.html ↩

  2. Annie Lowrey, “The Time Tax: Why Is So Much American Bureaucracy Left to Average Citizens?”, The Atlantic, July 27, 2021. https://www.theatlantic.com/politics/archive/2021/07/how-government-learned-waste-your-time-tax/619568/ ↩

  3. Annie Lowrey, The Time Tax: How the Government Wastes Our Time and How to Fix It, published August 25, 2026. https://www.publishersweekly.com/9780063442559 ↩

  4. https://www.ama-assn.org/system/files/prior-authorization-survey.pdf ↩

  5. https://www.ama-assn.org/practice-management/prior-authorization/fixing-prior-auth-nearly-40-prior-authorizations-week-way ↩

  6. https://www.acassessor.org/wp-content/uploads/2023/06/2022-23-Draft-Annual-Report-Letter-Size-6.1.23.pdf ↩

  7. https://www.spur.org/sites/default/files/2022-02/SPUR_Burdens_and_Benefits.pdf ↩

  8. https://www.ama-assn.org/practice-management/prior-authorization/only-1-3-doctors-trusts-insurers-prior-authorization ↩