I successfully defended the dissertation on January 9, with some minor edits made at the request of my committee. In the adrenaline of actually doing the presentation I forgot to record the zoom, which is a pity in that I made an effort to do a nice setup for it and my slides were very fun to design. Huge thanks to Miriam Nielsen for helping me with A/V—I would have been totally lost without her.
The PDF
For those with academic credentials, the dissertation is on ProQuest; for everyone else, I’m just putting the PDF up here. The main revisions ask from my committee much to my annoyance, was the removal of “overt jokes” (note that they did not explain what that meant or provide a list, which meant re-reading the whole dissertation to figure out whether a joke was subtle enough to pass muster). In terms of quality, it’s definitely something I wrote in approximately three months, so not my most sparkling prose. But there’s stuff in there that’s never been published before, which is fun.
Next step was basically a pivot away, sigh
Because I started a full time job (and a super-partial part-time job) in the final months of wrapping up my dissertation, I never really gave myself time to process being “finished” or reflect back on the experience. Also, the job(s) has (have) almost nothing to do with my dissertation (more on this “almost” in a moment) so it’s been hanging out on a far back burner of Stuff I’ll Get To Eventually.
That’s not entirely true: I gave a short talk on some of the dissertation research at Brooklyn Web Workers in the winter and presented some of it at a panel at AAG in March. The Web Workers audience was more enthusiastic about it but this may have more to do with the fact that it’s a meetup at a bar, not an academic conference. Also, my AAG panel was on the Saturday morning that marked the end of the conference, which is to say when most people had either already left or were too hungover to bother with morning sessions. I haven’t heard back from the editors of a journal special issue I submitted a draft to ages ago and I’m starting to suspect they just scrapped the whole article or something. There is some timeline where this research turns into a book, but I’m currently already behind on revising a new edition of my first book and can’t quite juggle two book projects right now. I’m also very behind on proofreading my transcripts, which isn’t a super-urgent thing but I’d rather do sooner than later, and I feel bad that I’ve failed to make time for this task in recent months.
Work overlaps
Ironically, I do think there is some useful resonance between research on geospatial software and some of the work I’m having to do now for my job, which is to say think about the political economy of big centralized large language model AI. (I know that I could just call it “AI” and everyone would know what I mean, but I’m not sure that’s a good thing.) LLM technologies are similarly challenged by the “parity with reality is expensive” problem, albeit in a more complicated sense of “parity” than, e.g., “is this road one-way or not.” Maintenance is necessarily an ongoing, perpetual project (although geo doesn’t have the problem of potentially corrupting its ground truth the way AI slop proliferation on the web does.) The tension between high utility potential and low profit margins is also pretty obvious, as seen in reporting about how AI tokens are outpacing the cost of just paying a software engineer a salary and the amount of debt financing going into the data center boom. And, of course, there’s the necessary subsidy of the state, and more particularly state violence. (A friend with a startup who’s spend a lot more time in the world of venture capital told me a while back that his “it’s a bubble” moment was when OpenAI added former generals to their board. “If they need inroads into government contracting, they already know they can’t survive on consumer and B2B,” he explained.)
A significant difference between LLMs and geospatial software: the industry hasn’t had an Overture moment, and I kind of don’t think it will. Competing on who has the best general-purpose or B2B-fine tuned models isn’t quite the same as competing on who has the better base map/routing app/POI dataset, but they are both products whose value arguably comes more from what they enable than the thing-in-itself. The actual code generated by Claude Code is fine but as far as I can tell not categorically better code than anything a seasoned engineer can write. The value is in the fact a company didn’t need to pay an engineer to write it and/or they got that engineer to work on multiple projects at once by using an agent instead, increasing productivity, which theoretically increases profits. But unlike geo, the business model of LLMs is charging for the thing-in-itself: the tokens, the compute. Imagine paying for turn-by-turn directions by the kilometer or restaurant search results by radius.
Obviously, this isn’t a one-to-one comparison. Geo is significantly computationally cheaper (though, to be fair, it’s not all that cheap), and the precedent for subsidized geospatial data was set in a very different time. It’s a lot easier to spin up open source geo software and data projects than it is to DIY an open source high-performing general-purpose large language model chatbot. But it is interesting to see so much work going into making competing products that low-key kind of all do the same thing after spending so much time writing about a bunch of big tech firms taking a very different collaborative open source approach upon identifying that competing on geo didn’t actually make sense. (I am not even getting into whatever the hell “GeoAI” is here, let alone world models, because I have but one wild and precious life and I would like to do other things with it, like commission articles about the history of Black indie media in New York City and make new drawings of manhole covers for my book.)
This domain name is really expensive so I might end up using this blog for other stuff under the broad umbrella of “placing” and “technologies”
One thing I really did like about doing this blog was that it kept me accountable to poking around and synthesizing various research rabbit holes into little updates, and I’ve been thinking that might be good to have for the projects I’m doing now. It’s kind of a hard pivot and I know I could just make a new blog, but also I just paid like $90 to renew this domain so it feels like I should get some more use out of it (I know, I also probably could get a better deal if I moved off of Gandi for domains (what are we using instead these days, folks?) but these gTLD companies are low-key just landlords of the poetic turn of phrase and they know it).
Some of the research work I’m doing right now is still germane to the theme of “Placing Technologies”, insofar as it is about the siting of internet infrastructure. One of those things is revising Networks of New York, the other is following the national conversation and localized conflicts over the massive data center buildout underway. It’s so strange to follow this shift in public understanding; I wrote explainers about data centers back in 2015 that were seen as fun niche reading and now it’s a mainstream media topic and midterm election issue. Apologies to those who came here for geo exclusively, but hopefully the other stuff will also be interesting to some of you.
Date
May 5, 2026
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.