Prologue

Just over a year ago, Christopher started talking about launching a newsletter about AI. This made perfect sense—he'd been experimenting with AI tools for geospatial work, and people in the GIS community kept turning to him with their AI questions.

Of course, I thought, we can handle, like, a monthly newsletter.

But then he said he wanted to do it weekly.

Sadly, we both have lots of job responsibilities that have nothing to do with putting Almost Entirely Human out into the world, so I was skeptical. Weekly is…a lot. That’s new content every 7 days!

And yet, here we are—celebrating one full year of this experiment. Thanks to everyone who has been following along, whether you are an OG subscriber or you just stumbled in here.

And, if you just stumbled in here, you are in luck, because I convinced Christopher he could take a week off building so we could highlight our favorite episodes—so you can make this into your reading list to get caught up!

Hollologue - Holly’s favorites

Episode 13 - How Many Calories are in this Cat?

Don’t worry, you don’t have to be into calculating your macros to appreciate this post! But, this episode is a great example of what kind of problems you can solve with AI—and the mechanics behind building these solutions. What I mean is, a year ago, when Christopher said "I built an AI to do this thing," I nodded along without really understanding what went into it. Now, I get the building blocks—like having an AI run the same prompt multiple times (as in this post) or using two different models to cross-check results (Honorable Mention to Episode 33 - Trust But Verif-AI). Which means I've gone from passive consumer to active tinkerer—I'm coming up with my own ideas for AI tools (see Episode 41 below) and even planning apps I want to build myself (perhaps you’ll see those in a future episode).

Episode 30 - Don’t be a Secret Cyborg

I think this one gets at the heart of how Christopher approaches AI, which I have really adopted as well. I have, in fact, become insufferable about sharing what I've been doing with AI. I very much love and appreciate human work and human writing and human artistry AND I love what AI has enabled me to accomplish. Let’s celebrate both!

Episode 21 - AI of the Beholder: Crafting a Civil War Book Cover

A real life example of AI + human artistry, and my first AEH guest post!

Episode 41 - Using AI to Take the Temperature of the Room

This episode, more than any other, really shows the evolution of how I think about using AI. I can guarantee that were I not digging in every week to Christopher’s latest AI shenanigans, I would not have invited an AI to monitor my Thanksgiving dinner conversation. Also, it feels like a full circle back to Episode 3 - Building a Thanksgiving Dish Assigner, where Christopher built a much simpler Thanksgiving-related AI app. Next year? Who knows what we’ll come up with.

Episode 38 - The Arbiter

Hands down, my favorite episode. (Wait, what does it say that my favorite is the one that neither Christopher nor I wrote?) Read this one for the best team-building exercise ever—there’s laughter, there’s an AI-existential crisis, there’s (eventually) dinner.

Christophologue - Christopher’s Favorites

Writing this every week has taught me a lot. First, I am a better writer! AI aside, I have never thought of myself as a good writer. Writing a newsletter every week, that at least three people will read (Holly, my Mom, my Dad—thanks, guys!) is stressful and nerve-wracking and it did not come easy. I spent the first few months stressing about what I would write each week. Would anyone care? Is it interesting? Am I getting my point across? At various points, the answer to all of those was yes, but it was also no. But I stuck with it (and so did all of you—thank you!) and I slowly got better at crafting a story and writing it. I'm no Jason Feifer, but I can manage, and I am going to keep working on it.

Episode 49 - How I Write Apps

This is my all-time favorite. I'm really proud of all of this one: the ideas, the pictures, the writing, all of it! Think, Engage, Test!

Episode 38 – The Arbiter

I think I loved doing this one better than writing about it. I spent all day after I released the Arbiter watching logs and looking at what people were doing to get the AI to betray me! (Note that is one appears on both lists, you really should go read it!)

Episode 7 – Dear HAL

I love this one because it is a perfect example of non-chatbot AI usage that can make a real difference for people. I really don't understand why more people don't use AI to analyze submissions from users; it has only gotten better since I wrote that post!

Episode 4 – Disposable Apps

I still believe in disposable apps, but this was also (about a year ago) when I demonstrated at Esri's Developer Summit that you could build an entire (simple) app using just AI tools. At this year's DevSummit, the AI is going to build a much more complex app and handle it all on its own. I'm just going to provide the commentary and the permission (and maybe I'll still get to play the hero if the AI can't do it).

What’s Next?

In a word: more!

• More livestreams. Monthly build sessions where I make things live and they may or may not break. Join us on the Third Thursday of every month!
• More conferences. I'm applying to everything! If you want someone to build an AI-powered GIS app on stage with a nonzero chance of failure, you know where to find me.
• New tools. Last month, dymaptic launched the Accessible Map Agent. Next week, you'll see what I'm calling "Project Alexandria!" I will also keep digging into how we're using Claude to control ArcGIS Pro. Plus much, much more to come! (say this in your announcer voice, it's better that way)
• More newsletter. Every week. Because that's who I am now.

Thank you for being here for Year One. Let's see what Year Two looks like.

Newsologue

• OpenAI's Responses API now lets developers build memory-enabled agents with minimal code. Instead of just returning text, your API call can now call functions, remember history, and coordinate multi-step workflows. This is basically offering the ChatGPT infrastructure as an API which can shorten the time and effort to build a real agent significantly.
• Mastra's Observational Memory is a new open-source approach to agent memory that outperforms RAG on long-context benchmarks while cutting costs by up to 10x. Instead of retrieving context dynamically like RAG does, it uses two background agents—an Observer and a Reflector—to compress conversation history into a dense log. Think of it like giving your agent a subconscious that's always taking notes. If you read Episode 48 about how LLMs handle (and lose) context, this is one approach to solving that problem.
• Claude Cowork came to Windows this week. Cowork is Anthropic's desktop agent that gives Claude access to a folder on your computer and lets it read, edit, and create files autonomously—like Claude Code, but for non-developers. The Windows release brings full feature parity with macOS, including MCP connectors and a new persistent instructions feature. It's available on all paid Claude plans and is still in research preview. If you've been waiting to try it because you're on Windows, now's your chance.

Epilogue

Holly wrote the first draft of this post, then I added my favorites and added a few things here and there. Then AI edited, and Holly edited and that was that. See you all next week!