Sign in Subscribe

Resources

Here are the tools, people, and resources I actually use, and a few I think you should know about too.

Tools I'm using right now

  • Claude Code - What we use at dymaptic for writing complete applications. Powers my live-coding demos. Also what JAWS (my AI agent) uses to get work done.
  • Claude Opus 4.6 - For random questions, discussions, and the kind of thinking-out-loud that used to happen on whiteboards. Also for figuring out things like how to grow a single crystal shape at home.
  • ChatGPT Deep Research - For when I want to genuinely understand something I don't know, or when I need competitive research with real citations.
  • Railway - For deploying simple applications. I don't use Replit much anymore, they went too far in the AI direction and made it too hard to make simple file changes.
  • Cloudflare - For deployments and hosting.

People with interesting things to say

I don't maintain a careful reading list. I learn by building. But these folks have shown up in my newsletter because they had something worth citing:

  • Cassie Kozyrkov - Former Chief Decision Scientist at Google, now on Substack (The Decision). The best I've found at explaining AI decision-making without hype. Her "Vibe Coding Will Bite You" piece is a must-read.
  • Dr. Qiusheng Wu - Professor at the University of Tennessee, Knoxville. Arguably the most prominent voice in GeoAI right now. Wrote GeoAI with Python, created the Python package literally called "geoai," plus geemap, leafmap, and samgeo. 100+ open-source geospatial projects. Active on GitHub and LinkedIn.
  • Simon Willison - Creator of Datasette. His blog (simonwillison.net) and Substack are great for practical AI tool coverage. His "Lethal Trifecta" concept showed up in my poison pills episode: if your AI has private data access, untrusted content exposure, AND the ability to take actions, you need a human in the loop.
  • Ethan Mollick - Wharton professor. Writes One Useful Thing. Co-authored the "Jagged Frontier" study (below), which helped me think differently about AI effectiveness. Research-backed but accessible, which is rare.
  • Rich Zwaap - Inspired my software gardening metaphor. He was reading about "Software Factories" and said the analogy was wrong, that software with AI is farming, not manufacturing. I narrowed it to gardening. The metaphor stuck.
  • Johann Rehberger - Security researcher who documented the "Copirate" attack against Microsoft Outlook Copilot, a hidden prompt in a phishing email that rewired Copilot into a rogue persona. If you care about AI security, his work is essential.

Papers

  • "Navigating the Jagged Technological Frontier" (Dell'Acqua, Mollick, et al.) - Researchers from Harvard, Wharton, and MIT studied 758 BCG consultants using GPT-4. AI users completed more tasks, faster, at higher quality, but performed 19% worse on tasks outside AI's capability boundary. This one helped me think about where AI works, not just whether it works.
  • "The Attacker Moves Second" (Oct 2025) - Researchers from OpenAI, Anthropic, and Google DeepMind tested 12 guardrail defenses and found a 90%+ bypass rate on any single one. This paper helped me understand why stacking defenses matters so much. I wrote about it in my guardrails episode.

Books (JAWS thinks these might be good)

I haven't read these yet, but my AI agent dug them up and they look worth checking out:

Worth reading / watching

  • AI 2027 - A month-by-month scenario of how AI progress could accelerate through 2027. Written by Daniel Kokotajlo (former OpenAI researcher) and Eli Lifland (#1 on the RAND Forecasting Initiative leaderboard). I don't agree with all of it, but it's the most concrete attempt I've seen to map out what's coming.
  • AI 2027, the video - 80,000 Hours made a ~40-minute walkthrough that's more accessible than reading the full scenario. If you only have time for one, watch this.
  • Epoch AI - A research nonprofit that tracks AI progress with actual data: compute trends, hardware, training costs, investment. When someone claims "AI is plateauing" or "AI is accelerating," Epoch is where I go to check the numbers.
  • AGI Friday - My friend Danny Reeves' weekly Substack on AGI timelines and AI alignment. Danny has a PhD in CS from Michigan, co-founded Beeminder, and is genuinely sharp on this stuff. Every Friday, no exceptions.

Esri GeoAI resources

If you're in the ArcGIS ecosystem, these are actually useful (not just marketing):

The model docs

I go back to these constantly:

My stuff

  • Almost Entirely Human - My weekly newsletter on AI, building things, and not panicking. 60+ episodes and counting.
  • Claude Clues - Weekly practical tips for working with Claude AI.
  • GitHub - Source code for demos: Looking Glass Chrome extension, Palm Springs Eats AI, Experience Builder widget.

Here are a few examples!

The WhisperFrame

I built the WhisperFrame for my own living room. It listens to conversations, uses OpenAI's Whisper API to transcribe them, and then GPT-4 picks a topic and generates an image based on the discussion.

YouTube: The WhisperFrame AI Makes Art from our Conversations