One of the ironies of my life at a startup is we’ll talk to larger companies that seem interested in partnering with us, but then they’ll get sort of quiet and say, “Look…we like you…but our master services agreement management system is…difficult. You may be…disturbed…by what we get up to…our contracts…our vendor onboarding…governance…” And then they look off into the distance on the Zoom call, and just mumble the word, “Process.” 

Little do they know that those words make us happy! It took me a long time to accept that I’m a bureaucracy fan. Talking through procurement? Yes. Creating a matrix of stakeholder sub-organizations? Gold. Open-source governance? SICKOS.png. I don’t talk about it too much because it makes people think I’m some sort of freak. What is wrong with this guy, that he wants to know how we use kanban to optimize our content production pipelines?

Why would someone be a bureaucracy fan? Aren’t bureaucracies bad? Not necessarily. I see them as unavoidable human artifacts—and deeply fascinating. You might study the real Roman Empire. Learning about how different FAANG companies do stack ranking? That’s my Roman Empire. Reading Miriam Posner on the cultural significance of SAP? Amazing stuff. More! (You want a good book to read? The Oxford Handbook of Megaproject Management—a sort of modern guide for what to do when your local pharaoh needs that new pyramid complex.)

There are lots of things to love or hate about AI, but the part that I keep coming back to—the part that doesn’t get VCs or Valley people super excited, as far as I can tell—is just how good it is at simulating bureaucracies and doing bureaucratic tasks. I would go so far as to say that you can predict how good an LLM will be at completing a task by identifying whether the task is one commonly performed in a bureaucracy or not.

Get an AI to write a short story? It’s likely to suck. But try this: “We manufacture ball bearings in every size. Create a PowerPoint that educates our team about the ISO-9001 certification process. Make it have zazzle!” The result will be a very, very good start.

Do you understand what that little lens flare could mean to the career of the middle manager who has to rope this thing together? No one at the ball bearing factory has ever seen anything like that before. I’m serious. Not only can I have ChatGPT make slides, but when I open PowerPoint, I can click the little icon and have it make even more slides. There is no limit, no rational conclusion, only an infinite plenty of slides. Is that what society even needs? I don’t make the rules.

Does that AI-generated slide really get at the fundamental essence of ISO-9001? Don’t ask that question. What’s wrong with you? What in the world would make you think ISO-9001, which is just one in the rich and abundant family of quality assessment standards known as ISO-9000, has a fundamental essence? I know we’re only supposed to work at jobs that align with our special natures, but in truth, a lot of life is plowing through a bunch of stuff that may or may not make sense because some bureaucracy demands it.

I would have high confidence using Claude or ChatGPT or their ilk to explore these subjects:

• The difference between ISO, IFRS, ECMA, and ANSI.
• Different org structures for marketing groups.
• Broad advice on supply-chain management software.
• The different kinds of mortgages.
• How to train a young sales team on a CRM.
• The advantages of being an S Corp vs. an LLC.
• The different frameworks for assessing quality in manufacturing.
• Which CMS might be best for a credit union to offer personal financial tips.
• What a provost does.
• The responsibilities of a not-for-profit’s board member.
• How to negotiate a parking ticket in New Jersey.

Would I trust their output? Not fully! You still have to validate everything. But I’d trust it more than I would trust it to diagnose a disease. My hypothesis is that it’s good at bureaucratic stuff because:

1. Large bureaucracies—like Google, or J.P. Morgan Chase, or the government of Canada—are the most powerful institutions on earth, and interact with millions or billions of people;
2. As a natural result, guidance for working with bureaucracies is incredibly abundant online, in countless publications, and in general “baked in” to culture;
3. AI companies—inadvertently—absorb that guidance and synthesize it into LLMs.

Bureaucracies are kind of the “background radiation” for huge swathes of public digital content. For example, every web page or ad or document about mortgages assumes an enormous banking system that is regulated and interacts with insurance products. LLMs compress all the connections between those ideas into big weird blobs of conceptual connection.

The upshot is if you ask an LLM about what tax strategy you should pursue for your llama farm, it will immediately tell you to “prepare a sample filled-out Schedule F with common llama farm expenses (feed, vet, shearing, fencing, etc.)” and then it will make a fake Schedule F for you complete with stud fees ($2,000). You should not send it to the IRS, but when you go to your accountant, or boot up TurboTax, you will know what you’re looking at. The bureaucracy will seem more normal to you. The world will be more navigable.

The same is true of code. LLMs are better at the bureaucratic parts of code. Claude can explain what code is doing. It is good at translating from one language into another, converting recursive functions into iterative functions, adding documentation strings, writing unit tests, and creating deployment scripts for cloud services. If you keep it there and write the very specific, custom parts of an app yourself, you’re going to have an amazing time. If you throw it all over the fence and pray the prompt does everything right? Crapshoot.

This is why I don’t know if vibe coding can deliver what it’s promising—because really good, unique, interesting software doesn’t emerge from bureaucracy. The awkward truth of the software industry, however, is most software isn’t unique or interesting. The majority of software on Earth has to: (1) capture data values; (2) store them; (3) periodically aggregate them into dashboard visualizations; (4) email the dashboards as PDFs; (5) keep doing it every day even though no one opens the emails. AI can really help here.

There’s this weird tension on this subject where the people who do the most commenting on AI are either technologists, who love the shiny robot, or on the other side, cultural experts—writers, artists, college professors—people who see creating things as an essential human calling, and are disturbed by what they see. But there’s a large cohort of Western-society bureaucratic workers who just gotta get that dashboard out the door or make the PowerPoint. 

I think at some level, those first two groups are going to be disappointed. AI won’t be as exciting as the technologists want it to be, because it’s actually just a weird database. Nor will it be as constrained and regulated as the cultural and academic cohort would like, because while their voice is loud, they aren’t powerful enough to stop what’s coming next.

But the absolutely vast behemoths of business and government that really do run the world? This is actually really great for them. They’d much rather have someone get some AI help making their ISO-9001 PowerPoint. That’s why they’re so all in on this stuff—because this whole technology is sort of a grease for the world’s biggest gears.

Does this mean the entire future is a Kafkaesque robot meta-bureaucracy where LLMs stochastically interact on behalf of disinterested humans? Probably. But I’m not sure it will look all that different from now. At the same time, the tools for traversing, understanding, and interacting with big organizations have never been better. It’s absolute boom times for a bureaucracy enjoyer like myself. 

I think back on all the giant multi-page Excel spreadsheets I had to fill out a few years ago—spreadsheets with Word docs inside them, as attachments that also needed to be filled out, knowing that no one would ever read them. They were mandatory to get to the next phase in the proposal process, even as I knew in my heart we would never be given the contract because they’d already decided to go with some part of Accenture. Now I have a tool that could make that process so, so much better—something that would do the meaningless work for me, so I could just approve it and move on. I’m going to enjoy it while it lasts.