📚 Neat Stuff I Read in March 2026

On taxes, saving money, and how AI can expand the scope of your tasks. Also: merchant rulers, the history of the Caucus region, and the unique historical advantages of Scotland and Florence.

This month’s link roundup is coming to you a day late because I personally avoid the internet on April Fool’s Day. Just about the only joke I appreciated was RobRoy subtly adding funny typos to the channel names of the nerdy Discord server he runs… which was much outweighed by needing to triple-check every nerdy science thing I ran across on twitter.

Anyway, March’s list has less ‘productivity’ and more ‘ai’ information, but I focused less on tooling and more on figuring out the impacts of all these new ‘productivity boosts.’ I still found plenty of time to read about neat history and science stuff too. I also picked up the latest fantasy novel from my favorite authors, the husband-and-wife team of (Gordon &) Ilona Andrews. It’s a portal fantasy where a woman from Texas — “Maggie the Undying” — ends up in the world of her favorite fantasy series (which, much like my favorite childhood fantasy series, tragically lacked the expected third book in the trilogy), and uses her knowledge of events (past and future!) to try and prevent all the nasty stuff she knows is coming down the pike for her favorite characters.

Those are, by the way, Amazon affiliate links, I earn a small commission if you click through and buy, blah blah it’s 2026 you know the drill.

Anyway, on to the rest of what I read in March!

Productivity & AI

• Hana Lee Goldin, MLIS, a librarian by training, put together a handy bunch of Google tricks. I have pretty good Google-fu already, but I learned a few things from this one. Might be worth teaching some of these tricks to your AI agent, too.

• Katie Parrott at Every wrote about how AI was supposed to free her time but consumed it instead. Basically the hot new phrase is “Jevons paradox” where cheaper tasks just make you attempt more of them. I found it extremely relatable because I often “save time” with Claude only to immediately expand the task into things I wasn’t able to do at all before.

• Max Woolf, historically an AI skeptic, wrote a detailed walkthrough of trying AI agent coding with Opus 4.5 for Rust and Python. It reliably produced correct, performant code, and I trust his report specifically because of his earlier skepticism. I also appreciated his aside that he intentionally makes his writing voice more sardonic to fend off AI accusations.

• Daniel Phiri wrote about why filesystems are having a moment in the AI agent ecosystem. He also flags finding that context files can actually hurt agent performance when they’re too long, which tracks with my experience. I do wonder how long it’s going to be like this; there are a lot of benefits in abstracting this stuff away from “normal” users (see also: Google Cloud).

• Anna Tong and Rashi Shrivastava at Forbes reported on how Cursor is scrambling to stay relevant as Claude Code and Codex eat into the code-editor model. We’re currently in an age where companies like Anthropic are “subsidizing” the “enterprise” plans (200$/mo Claude Code is like $5k in API costs). So Cursor’s working on building its own coding models on top of DeepSeek and Qwen. Personally I am keeping a close eye on the “cheap, lightweight specialist” AI vs “expensive, big, generalist” AI race. For example, someone figured out that dedicated customer service chatbot models outperform general-purpose ones.

• Cesar Hidalgo wrote about what happens when you loop an AI paper-writer with an AI reviewer. The resulting paper kept getting longer because the AI reviewer never exercised “strategic forgiveness” the way human reviewers do, which is to say that they kept hammering instead of considering a response “good enough” and getting tired of asking for revisions. My own experience is the opposite — AI back and forth results in the agent trying to get out of fixing anything because “it’s probably just a flaky test, no big deal” when the whole app is nonfunctional :P

• Jeffrey Emanuel wrote a practical guide to feeding your tax return to an AI, decomposing it into context-sized chunks to find deductions and spot errors. Some of the specific techniques are already outdated since context windows have grown, but the core idea holds up. I used AI to help with some tax forms this year and it saved a real chunk of time, even accounting for a thorough manual check afterward. In point of fact, it made fewer errors than my actual tax professional, who missed a $49 entry during data input. Patrick McKenzie, who I trust completely about financial matters, as found AI helpful for tax things too.

• Nir Barak wrote about rewriting JSONata as a pure Go library with AI in seven hours, going to native Go and saving $500K/year. The main thing was having a well-defined spec with many hundreds of existing test cases to validate against, but this is interesting to me mostly because it’s a good example of a really beneficial way of using AI rather than just shipping faster slop or, you know, building dashboards.

• Mario Zechner (who built pi, a lightweight and extensible coding harness that I’m coming to love) wrote about slowing the fuck down with agentic coding. His argument is that autonomous agent swarms compound tiny errors at a rate humans never could, because agents don’t learn from mistakes and there’s no bottleneck to slow the rot. Agents cargo-cult bad architecture from their training data and each agent only ever has a local view of the mess it’s creating. It’s very tempting to completely refactor my vault with agent swarms, but thus far I have avoided giving into temptation because of precisely this.

• Dan Lewis wrote a measured breakdown of the fault lines in the AI economy, assigning percentages to six plausible scenarios ranging from “flop” to “structural post-work.” I appreciated the historical grounding in stuff like the Industrial Revolution and the Great Depression and the fact that it seemed reasonable instead of hyperbolic panic or dismissiveness.

• Weco is an autonomous code optimization platform where you hand it source code and a performance metric, and it iteratively proposes and tests improvements. I share this not because I expect anyone here to purchase their services, but because the examples page is an unusually good look (compared to, for example, twitter hype) at where AI is really paying off. For example: a massive improvement on detecting X-ray abnormalities and a big improvement on cleaning up images in preparation for OCR.

History

• This month I read Stephen R. Bown’s Merchant Kings, which covers six monopoly trading company leaders who ended up as de facto dictators from 1600 to 1900. It was a really interesting book for 95% of the way, and then went off the rails at the end when he started moralizing, which involved a lot of contradicting his own words and speculating in the most uncharitable ways. I still recommend it for people curious about how obsessively driven mercantile men helped shape the world, but I’m irked at the epilogue.

• Paul Graham wrote about the brand age, tracing how Japanese competition forced the Swiss watch industry to shift from craftsmanship to brand identity. It was a bit surreal because I had just written about Japanese artisanal watchmakers in my review of The Perfectionists by Simon Winchester.

• Jake Hubbert wrote a paper for BYU about how Islamic visual culture impacted medieval Georgia’s religious and secular art. It traces Persian portraiture techniques through monastery frescos, illuminated manuscripts, and fortress architecture from the 7th to 14th centuries and was a pretty neat look at a country that I have learned remarkably little about before. I got interested in Georgia because a team is studying the fortress of Dmanisis Gora in Georgia to figure out why Caucasus communities weathered the 12th century BC crisis better than the rest of the Middle East. The site is unusually well-preserved because of how flat it is, so they’re hoping to find stratified layers from right around the collapse.

• Maxi Gorynski wrote a long, affectionate essay on nearby Armenia, covering everything from its status as the first nation to formally adopt Christianity (301 AD) to its Bronze Age archaeological record, which includes the world’s oldest shoe, wagon, and wine-making facility.

• Rune Iversen published a classification of over 600 engraved “sun stones” from Neolithic Bornholm in Antiquity. The stones were deposited around 2900 BC, probably in response to a volcanic cooling event that darkened skies and wrecked harvests across the northern hemisphere. It’s pretty interesting how consistently people get more religious during times of stress, even five thousand years ago.

• Lucy Worsley wrote a walkthrough of the surprisingly dramatic history of candles, from tallow-dipped medieval rushes to the absurd logistics of lighting the Hall of Mirrors at Versailles. It’s not quite Salt or Cod, (or even Coal, which I reviewed back in February), but it’s a nice look into how a mundane object can be a window into class, labor, and the politics of who got to see after dark. I particularly enjoyed the pictures.

• Tomas Pueyo wrote about why Florence specifically sparked the Renaissance, tracing a convergence of geography, political independence, wool-trade wealth, and proximity to Roman ruins. A ruined cathedral plays a surprisingly big part in the story, and it reminded me a little of other arguments I’ve seen about why it had to be Britain for the Industrial Revolution. His angle for carying about Florence is figuring out what makes places like it and Silicon Valley so special, so we can replicate the magic. I’m not sure that’s doable on command, but it was still an interesting story with lots of interesting maps and graphs.

• Anton Howes wrote about why Scotland punched so far above its weight from the 1740s onward. It’s a convincing case that Scotland’s real advantage was its financial system: Scots law let unchartered banks grow much larger than English ones, overdrafts were invented in Edinburgh in 1728, and even artisans without land could earn interest. Glasgow’s merchants used that capital edge to dominate the tobacco trade, and when the American Revolution wiped out their Virginia warehouses, they were well-capitalized enough to just pivot to Caribbean sugar. It’s a good companion piece to the Florence article above.

Science

• Ryan Burge (the “Graphs About Religion” guy) looked at why belief in the afterlife persists even as church attendance drops. 88% of Americans believe in a soul, and the youngest adults in the survey actually bounced back toward higher belief in heaven and hell compared to millennials. The concept of “religious residue” was a new term for a phenomenon I’ve definitely come across before — you can leave the church, but spiritual frameworks stick around in your psyche longer which is how we end up with the concept of “cultural Christians” and “secular Jews”.

• One neat thing about chimpanzees that I never noticed when I visited them in zoos... you can’t really tell where they’re looking. Scott Travers went pretty deep on why humans are the only primates with white sclera, and the answer gets into self-domestication (which you may remember I touched on extensively in my article about how and murder were critical to the domestication of Homo sapiens sapiens), the evolutionary value of eye contact, and the genuine tradeoffs of making yourself more readable to everyone around you.

ICYMI

• I reviewed Steve Brusatte’s excellent book The Rise and Reign of the Mammals, which covers hundreds of millions of years of mammalian evolution. I had a lot of fun with the deep-time ecology stuff and learning about how important teeth, jaws, and ears ended up being.

• I wrote about geldings and the “natural” social order of horses, starting from a throwaway line in an archaeology paper about Mongolian herds. It’s part horse history, part philosophy of domestication, part reflection on the value of ‘unproductive’ males and the nature of ‘natural’ as a concept (see what I did there?). This one got a more likes but fewer comments than I expected.

• I also wrote about why I built two custom dashboards instead of using off-the-shelf analytics, and what I learned about the difference between data you look at and data you actually use.

Stay tuned for my upcoming review of The Cold Start Problem: How to Start and Scale Network Effects by Andrew Chen, an article about different ways organisms change their phenotypes, and a look into the benefits of diversity for large empires.

This month I’m reading Hard Landing: The Epic Contest for Power and Profits That Plunged the Airlines into Chaos by Thomas Petzinger, Jr. and it’s fascinating so far. Next month’s book is slated to be How Africa Works: Success and Failure on the World’s Last Developmental Frontier by Joe Studwell. I’m planning another visit to Harpers Ferry with my son, and looking forward to seeing Project Hail Mary (which everyone I’ve talked to says is great, a relief because I loved the book) with my husband over Spring Break.

And, as always, if you read anything particularly interesting last month — or end up reading anything above and want to discuss — please leave a comment or shoot me an email :)