📚 Neat Stuff I Read in January 2026

Tips for improving your AI-monitoring intuitions, intellectual endurance, and grasp of power-brokering institutions. Also: neat parasitic worms, a military history book club, & Bronze Age Collapse.

January was pretty wild for me. I finally dove deep into Claude Code and it’s been incredibly useful for both work and hobbies. I’ve spent most of the month bookmarking resources and experimenting with automation, so this month’s roundup is heavy on productivity and AI tooling, but don’t worry I still found plenty of neat history and science stuff too. Some links are Amazon affiliate links, but as always, I value my reputation more than double-digit dollars and am not trying to sell you irrelevant stuff I swear.

Productivity

• Ian Leslie article about the Stamina Gap was simultaneously terrifying and validating. He talks about how reading long novels builds cognitive endurance, and how sustained concentration skills are becoming increasingly rare (some would say elite) skills. I am good at both of these things, but this sort of thing is part of why I’m so determined to read one nonfiction book a month this year. I do not want to lose the ability to concentrate on hard problems, and the best way to keep my mind sharp is to exercise it.

• Brooke shared her approach to curating a journaling ecosystem because she was worried her “phone was making her stupid.” She ended up replacing doomscrolling with analog journaling to reclaim creativity and intentional time.

• John Psmith (who, along with his wife, writes one of my top five favorite newsletters) reviewed How to Solve It by George Pólya. It distills four core problem-solving steps (understand, plan, execute, reflect) into useful heuristics.

• A philosophy undergrad started daily workouts to avoid “becoming lost in the world of signs and forget the things they signify.” As something with an undergrad philosophy degree who struggles with exercise, I’m really trying to convince myself to think this way.

• A PhD candidate in economics at UC Berkeley argued that ideas aren’t getting harder to find, and research productivity hasn’t fallen. Apparently firms produce as many breakthrough patents per R&D dollar, but market/allocative inefficiencies mean those ideas no longer translate into broad productivity gains. This feels like something worth paying attention to from a “caring about improving the world” perspective.

AI Tools

• Lance Martin shared how he steers Claude Code using a /TODO command that finds #TODO(Claude) comments in his code and builds a plan. I do something pretty similar although I usually just do %%CC: note %% as my formatting. Anyway, more interesting was how he discussed “Fingerspitzengefühl” ― the intuitive feel for steering agents via Plan Mode, file referencing, and targeted TODOs. It’s the first time I’ve seen the term in the wild since I wrote my review of Certain to Win by Chet Richards. I genuinely think that developing this “intuitive sense” of what’s possible is the most important thing I can be working on right now, for reasons Martin Alderson outlines in his article about how two kinds of AI users are emerging.

• Boris Cherny (Claude Code’s creator) discussed how he uses it. Eyad shared a complete Claude Code tutorial. Basically, folks emphasize the value of thinking through things, and iterating over careful plans with curated context before sending prompts. Use compacting/clearing, run instances in parallel locally, use slash commands, subagents, offer verification methods, and provide examples for when the AI gets stuck. I’m not a programmer but I’ve personally managed to run 5 terminal “channels” in parallel and it felt pretty natural, like maintaining a couple of conversations in Slack or Discord at once. AI is finally worth using for a wide variety of white collar tasks; it even helped me with my taxes (yes of course I double-checked, I double-check my tax professional too).

• Klöss wrote a guide on how to improve at prompting. Don’t treat prompts as polite requests, treat them as code commands. Ordered context, clear constraints, set role definitions, structured output formats, and canonical documents all help make prompts more effective.

• Ryan Carson explained how to make your agent learn and ship while you sleep ― he suggests a nightly loop where the agent reviews threads to extract learnings into a controlling markdown file. I do a variation of this; I add to my memory file throughout the day, and have a “good-night” prompt that checks on a couple of places I update and compiles the new information where it belongs.

• Shruti spent a week researching Clawdbot, an autonomous AI that runs on a persistent server (like a Raspberry Pi or Cloudflare worker, although some people are using a Mac mini). It’s intended to execute tasks 24/7, but requires significant setup for complex workflows and can get pricey if you use APIs instead of riskier choices giving it its own CLI coding tool subscription. I got one working on a $5/mo Cloudflare worker but it was a pain and I’m nervous about the security implications, so I haven’t leveraged it much beyond learning how Puppeteer works (badly for what I need :P).

• Steve Yegge introduced Gas Town, an orchestrator that automates and swarms massive Claude Code agent workflows using tmux and a heartbeat system to run persistent, durable “convoys.” It sounds like industrial-grade chaos for advanced users only, and I have not tried it, but I’m seeing this “heartbeat” thing so much that I did implement a couple of cronjobs for repeated tasks I don’t want to have to personally trigger. I think Yegge is right that this is probably the future, so I’m trying to stay aware of it even if I’m not quite ready to dance on the bleeding edge.

• I’m more at the level of… did you know you can stash your prompt in Claude Code with ctrl+s to save drafts and auto-restore them? Here’s a straightforward guide to the bash commands LLMs use most often, so you can understand what they’re doing a little better.

Science

• Apparently parasitic worms use static electricity as a tractor beam. Researchers discovered hookworm larvae can harness static charges to leap onto hosts. The worms generate charges through movement, then ride the electric field gradient to their target — a previously unknown mechanism in parasites. Probably the neatest thing I read last month.

• My son and I have really been enjoying this “Insane Biology” series on YouTube. We started out listening to the audio stream neat video about octopus biology on his beloved Yoto (here’s a neat DIY version that is always-offline). Octopuses have 500 million neurons (2/3 in their arms, not their brain!), can change color in 200 milliseconds despite being colorblind, and represent a completely independent evolution of intelligence.

• Dan Davis History has a great video on how ancient salt mining transformed prehistoric Europe. Neolithic and Iron Age communities centered settlements around salt springs, using ceramic “brickage” and evaporation to produce salt.

History

• Harvard Magazine revisited a 1884 cannibalism at sea case that still anchors law school ethics debates (which were my favorite part of law school). This case basically reshaped Anglo-American law by rejecting necessity as a defense for killing an innocent to save others.

• Santi Ruiz looked into governance lessons from the Constitutional Convention and determined that small working committees plus secrecy and careful press signals were key to producing compromise. It reminded me a lot of L. E. Modesitt’s points in his gaslamp series The Grand Illusion. They’re more political theory than rollicking fun, but I still liked them and you might too.

• In January I finished reading 1177 BC by Eric Cline (who is very nice on Twitter and has very fun ties), which explains how the Sea Peoples are probably not really responsible for one of the first multi-civilizational collapse. I respect Cline as a scholar, and really enjoyed Chapter 5, but it’s more of an introduction to the region/era than a deep dive. Bret Devereaux happened to write about the Late Bronze Age Collapse recently as well, and in some ways I preferred his version because it was shorter, more targeted, and had much better maps.

• Naval Gazing is starting a book club read-through of Morison’s The Two-Ocean War, one chapter per week starting Feb 4. If that’s your sort of thing, go join them, Bean is great.

Modern Instutitions

• Matt Lakeman is my favorite travel blogger bar none, and his Notes on Afghanistan (which involved several Taliban encounters) were as fascinating as always. He has great photos, great book recommendations, and lots of great stories.

• This deep dive from Konstantin Asimonov into how to address a Russian came up during a discussion with some of my Russian-speaking friends. It broke down details about first name, patronym, family name quirks and discussed different ways of signaling precise levels of familiarity and respect.

• Years ago, Tony Kulesa wrote about how Tyler Cowen is the best curator of talent in the world because he consistently identifies undervalued high-potential talent across a wide variety of fields. I have no interest in VC, but I do care about hiring practices and surrounding myself with awesome people, and I really liked the way the article explained Cowen’s unique perspectives on this. I suspect off-the-wall methods like this will end up more popular in the coming years as the AI-vs-AI resume battles spiral out of control.

• Packy McCormick had a similarly fascinating profile of a16z as power brokers. The bet is that owning capabilities and narratives lets them invest bigger and shape industry norms. It’s not the sort of thing I want to do, but it’s definitely the sort of thing I want to understand in the same way I care about weather patterns and local school board trends.

• For similar reasons, Samuel Days reviewed Why Nations Fail, and declared it “an above-average book, which I consider damning with faint praise for a pair of Nobel Prize winners.”

ICYMI: My January Articles

• I wrote a review of The Perfectionists by Simon Winchester, which was a really good history of precision engineering. I enjoyed the guy who was so obsessed with iron that he built an iron desk, and loved learning how jet engine turbine blades operate at temperatures higher than their melting point by using ancient Greek-style lost-wax casting combined with microscopic cooling holes.

• How Locusts Cause Famines and Shaped the Middle East is probably one of my more interesting academic deep dives, judging from what friends have said. The article discusses swarms from biblical Palestine to 1874 Nebraska to modern East Africa. I focused mostly on Aaron Aaronsohn, the WWI-era agronomist who fought locusts in Ottoman service and because of that experience ended up spying for the British out of Zionist fervor. He is probably the basis of Liet-Kynes from Dune, except his problems were trickier.

• Claude + Obsidian Got a Level Up was my most popular article in ages. Like I said in the intro, Claude Code finally clicked for me. I connected to a Telegram bot so it can send morning reports, fixed my RSS plugin for Obsidian, set up automated fine-grained git commits, have a much-improved research flow, and refined how I process notes from books I’ve read. Basically, I’m not trying to do things faster, I’m trying to do them with less attention so they actually get done. The article includes anecdotes from my offline/analog personal life that help show how how I think lets me leverage AI more effectively.

• The most popular link from last month’s edition of Neat Stuff I Read was nobody knows how large software products work by Sean Goedecke. If you missed it, check it out!

Stay tuned for next week’s review of Coal: A Human History by Barbara Freese and as always, please do share this or any of my articles with a friend who might be interested :)