Kyle Walker

Working on H3 integration into freestiler, my R/Python vector tiling tool. The new feature uses @duckdb internally to dynamically aggregate points into a multi-layer hexagonal tileset based on some function (summary stats, or your own custom SQL). Shown here: the 12M US groundwater wells dataset. Hexagons are styled by modal category and H3 resolution changes by zoom level in @Mapbox, giving way to individual locations at zoom 10.

A data map turns location into a variable. Patterns start to show across space. The new #DataViz Weekly leans spatial, showing what maps can add when data has a geographic side: 📍 146M U.S. jobs 📍 Sahel violence 📍 Rio's sister cities 📍 Smoking endgame anychart.com/blog/2026/05/2…

The distribution of AI is incredibly uneven. Some individuals have fully autonomous agents working on their behalf while the general population barely knows what chatgpt is. AI-native companies are building agentic workforces while big corporations are deploying copilot. It's a modern day gold rush for those with agency and curiosity.

@rauchg Texas ag / wildlife exemptions and property tax protest guidance. exempt.land Claude 4.7 and Codex 5.5

People give Gemini a hard time because they only think about AI through the lens of agentic coding Gemini has been, and still is in my opinion, the #1 model for agentic document extraction and document understanding Which I think is one of the most important use-cases for concrete ROI from AI in tons of industries I've built multiple harnesses with Gemini to do doc review / paperwork and it really shines

@maksh3adr00m the workshops are here: walkerdata.gumroad.com/l/all-workshop… if you mean the skill - no that's not published but I'd recommend feeding the workshop QMDs to Claude and seeing what it can do!

@kyle_e_walker Are they online somewhere?

I'm retiring my programming workshops tomorrow, but they still can be very useful with modern AI tools. I used Claude to build a skill from my 11 Census / mapping workshops, then taught an OpenClaw bot that skill and gave it the 11 tutorials. I gave it instructions via Telegram to do a senior housing analysis using the skills it learned in the workshops. It prepared an analysis for me with interactive maps and Census data exports using what it learned. Even though coding tutorials are static, they can provide differentiated context for your AI agents and help you learn + produce new results. Grab the full catalog of workshops for only $99 - they go away forever tomorrow: walkerdata.gumroad.com/l/all-workshop…

More about the data: census.gov/data/experimen… Map it at the tract level yourselves: gist.github.com/walkerke/71e74…

I grew up in Oregon without air conditioning, it seemed quite normal at the time, we just pulled out the box fans when it got hot in the summer. Fascinating to explore the patterns in this new Census data - lots of areas around Oregon (and the rest of the US) where AC prevalence isn't as high as you might expect. My home Census tract in Corvallis is now at 94.7% with AC (my parents eventually got it too)

Final reminder: today's the last day to grab my Census / mapping workshop catalog before the store shuts down tonight. Lifetime access to 25+ hours of recordings, 11 custom tutorials, $99. I really appreciate all of you who have bought it this week! walkerdata.gumroad.com/l/all-workshop…

Mapped: the % of occupied housing units without air conditioning by Census tract. Uses the new @uscensusbureau LACE dataset. Map the data yourselves: gist.github.com/walkerke/71e74…

@chadbr They do! Was fun putting this feature together.

@kyle_e_walker They have great data 👍

New Mexico production data is live in the Wells Intelligence app and Clearfork Wells API. Updates include: - Oil, gas, and produced water history for NM since 1993 - Fast mapping, well selection, and decline curves for NM wells - PLSS survey area search - API and CLI access to NM data for your agent workflows Same pricing. No sales call or contract. wells.clearforkintelligence.com

@brunomnetoo yeah I still get nervous about agents working on main (unless it is a very tightly scoped mapgl task) - worktrees are the best

@kyle_e_walker Worktrees are the way to go. A well defined claude.md Auto mode was definitely a gamechanger/timesaver

Two months later, I have to admit that I was wrong about this. Auto mode is a beautiful thing if you use it correctly. If you: - Have a well-structured codebase with detailed guidelines and workplans (I use .md files for this); - Are carefully tracking with Git and telling your agents to commit regularly; - Have a test suite and well-designed regression tests; - Have clear guardrails around interacting with prod / the main branch (I use Git worktrees); you can trust your agents to do a whole lot. You're essentially managing an employee (or a team). If you manage your team well, they can do a whole lot independently. Now when I don't do those things I suggested... it's anything goes

@MadEngineering thank you! Gumroad takes a cut, but worth it as it handles hosting of the materials + payment processing

@kyle_e_walker Do you get 100% or the money from the sale?

@DataShaun Tons. My workflow: - I have it read all the chapters of my book (which is helpfully written in Markdown) to learn my voice - Claude writes a first-pass documentation - I go through and edit out awkward language Huge time-saver.

@kyle_e_walker Has it helped you with documentation?

People rant about “slop” but my open source packages have (or will soon have) full test coverage now and I’m making more progress on clearing out issues. I did not have time for this before Claude Code / Codex

@AJAllchin it's cliche but many critiques (mine included) boil down to "skill issue" I wasn't good at auto mode two months ago. Now I am, and I'm getting awesome results

@kyle_e_walker Yeah, I was surprised at your initial stance. Good to see you've come around :)