Every once in a while, I share an update of what I’m doing with AI. I believe in transparency when it comes to AI use. It’s not just good practice; it’s how we learn about AI and how those around us make decisions about it. I would love to see more conversations like this happening in schools, especially between students and educators.
So, I’ve put together this snapshot of what my AI use looks like right now. This is not advice. It is information. What works for me may not work for you, but I hope learning about what I do helps you be more curious about AI.
Day-to-Day Habits
I think it’s important to start by saying that I probably don’t use AI as much as you think I do. I interact with chatbots and other generative AI tools daily, but it’s usually in a few small chunks over a day, not for long, extended interactions unless I’m learning new skills and tools.
For example, if I’m building something with AI (training a bot, vibe coding an application, managing an agent to complete tasks… more on that below), then I’m spending an hour or more of focused time working with whatever the specific tool is. I really enjoy it. The prompting and troubleshooting and coaching of AI is immersive in a way that reminds me of when I was an instructional designer building online courses. It feels creative and rigorous.
The majority of my interactions with AI, however, are as a process assistant: I’m working on my own or with another person, and I dip in and out of AI tools for brief interactions. Claude, Gemini, and ChatGPT have become my go-to platforms for this (I pay for “Pro” or “Plus” subscriptions to all of them).
I usually take a “human first, human last” approach. These interactions almost always begin with inputting my own ideas and uploading content that I have created or curated. I want the bot use knowledge that I provide, not just pull it from the ether of its dataset. Plus I feel more confident evaluating and responding to the output. That’s the “human last” part: I vet, edit, and approve actively and ruthlessly. I reject far more AI content than I actually use.
For example, if I’m using a bot for workshop planning, I’ll share slidedecks I’ve used in the past and ask the bot to find discrepancies or connections or pathways for new ideas. I often push the bot to reimagine my plan, not merely adapt it. I ask it for multiple versions so that I can compare and contrast. I ask it to present me with scenarios I could face in terms of audience reaction or questions, or imagine ways a session could go awry and offer contingency plans. I take the output from one bot and copy/paste it into another, asking the second bot to critique or improve the output from the first (still one of my favorite and most effective use cases of AI).
Basically I’m using bots as versions of coworkers. To use a lot of buzzy workplace words, sometimes we scrum, sometimes we red team, sometimes we do design thinking, etc. I have been regularly surprised by how good the suggestions are. Once, without really thinking, I said to Gemini, “Wow... I have to say, I’m really impressed by these ideas.” It responded, “I’m so glad to hear that, Eric!” and I decided to take a break.
These process interactions rarely last more than 10-15 minutes at a time, and they have revealed to me how much better these models have gotten. Last month, I was unhappy with Claude’s feedback on a section of a talk I was preparing on learning design, and I finally just wrote, “This isn’t working for me. Hard to say why.” Its response was that it was going to be “more provocative,” and it offered up alternate plans based on conversations we had about change management research the week before. It was a reminder of the sophistication of these tools. They are no longer flummoxed by vague prompts, are skilled at using their knowledge of you to refine output, and can decide to veer from the focus of the conversation to look at a problem in a different way.
Importantly, I have given these bots guidance on how to interact with me. I often toggle to the “thinking” or reasoning model for more sophisticated output. I’ve activated Claude and ChatGPT’s “memory” feature so that they can search and reference our previous chats. I’ve also set the following settings in both bots: “I like concise answers that are candid over validating. Whenever possible, create tables and other visual organizers to help me digest info. Do not over-analyze. Give me initial thoughts and let me follow up.” I want to influence the default way I interact with these bots. I am regularly surprised by the number of people I meet who pay for these tools but have not spent any time customizing them.
Building my own “Workspaces.”
I have, over time, built workspaces in chatbots that I use as hubs for specific topics. For example, I have a notebook in NotebookLM that has all the articles from this Substack as well as dozens of slidedecks from workshops I’ve run over the past two years. I dip into this “Eric Hudson Database” when I want to remind myself if/how I’ve covered a topic, or if I want to incorporate something I’ve written about into a presentation (or vice versa). If I am working on something new, I will often open this notebook and simply ask what, if anything, I’ve already created on the topic.
The equivalent spaces in ChatGPT and Claude are “projects.” I’ve started using projects as places to gather and work with research: I have a project in Claude on change management. I have a project in ChatGPT on the impact AI use can have on critical thinking and learning. I upload resources as I find them, and I use the project-specific chat to ask questions or to help me pull references for my writing or workshops.
I’m Back to Making Bots
The last time I wrote an update like this, about a year ago, I had stopped building custom bots like GPT’s or Gems. That has changed. When I’m facilitating professional learning—AI-focused or not—I have found it really helpful to offer participants the opportunity to work with a custom bot I have trained. When I work with middle leaders on feedback, there’s a segment where they use a bot I’ve created to discuss and prepare for a challenging feedback conversation. When I work with educators on learning design, there’s a segment where they work with a bot I have trained to analyze and adapt one of their own assignments.
This has solved a real challenge for me: how do you introduce an element of personalization into a setting where there are dozens, maybe hundreds, of people in the room and only one of you? This has made the work of building and training bots (and it is a lot of work) worthwhile to me.
Tiptoeing into Vibecoding and Agentic Capabilities
I first wrote in depth about agentic capabilities in chatbots a few months ago. I’m still pleased with Gemini’s ability to perform simple tasks in my Google Workspace like create spreadsheets and forms based on information in documents and create/manage calendar events and generate reports based on searches it does in my Drive. None of this is revolutionary: a lot of longstanding workflow and project management tools have similar capabilities. But, the ease of doing it in natural language with a bot is appealing.
For now, I’ve walked away from agentic browsers like ChatGPT Atlas because so many of the online activities I would love to delegate to these tools would require sharing a credit card number, login credentials, or other private information that I don’t trust they can keep secure.
I know that the real potential is in tools like Claude Cowork, which I have on my computer, and I’ve run some successful early tests in getting it to clean up my hard drive, do some analysis of big spreadsheets of qualitative data and interview transcripts, and remix and reformat slidedecks. But, as Ethan Mollick makes clear in his overview of Claude Code, this barely scratches the surface of what we can now delegate to AI. And, as Stephen Fitzpatrick notes in his own exploration of Claude Cowork and Moltbook, we have moved out of the “human in the loop” era and into the “delegation without comprehension” era of AI. As much as I think about AI augmentation and automation, I now think about AI autonomy.
AI in my Personal Life
If I do find myself using AI outside of work, it’s usually very practical. I use voice mode for house stuff: ChatGPT talked me through how to replace a water filter in my refrigerator and how to troubleshoot my heat pump when it wouldn’t activate. Gemini helped me plan a pantry in an awkward under-stair space in my basement. Sometimes when I’m driving, instead of listening to music or a podcast, I’ll speak Spanish with ChatGPT just to practice (I already know Spanish, which is key. If I were learning a new language, I would use a learning tool like DuoLingo).
What I Don’t Use AI For
I don’t use AI for entertainment. I’m not interested in having it create stories for me to read, images for me to share for laughs, or songs to listen to. I can’t keep up with all the human-created content I’m supposed to be enjoying.
I don’t make videos using AI. The energy consumption cost compared to the poor quality of most of the output makes it not worthwhile to me.
I don’t ask AI to write for me. For example, I have written every single word of my Substack articles. I genuinely enjoy the process of putting these articles together, and I’ve no interest in delegating that work to AI. I still write all of my emails myself, which increasingly makes me feel like a dinosaur.
I don’t use AI for companionship. I would rather sit in silence and stare into space than engage a chatbot in a personal conversation.
AI is Infrastructure
As I hope these examples illustrate, the novelty of AI in my life largely has been replaced by the utility of AI in my life. Of course I’m still exploring new capabilities (it’s literally my job to do this), but more often than not I’m just using it. It’s functional for me, not magical.
This is, of course, the goal of the companies that own these frontier models. They are working to make AI so ubiquitous and frictionless that it becomes impossible for us not to use it. But, as I hope is also clear from my own use, I am resisting letting AI become my only or my default mode of working, thinking, or creating. As Marc Watkins wrote a year ago, we should think of AI as being unavoidable, not inevitable. We should not be foisting this technology on each other; instead, we should insist on making this infrastructure visible to each other so that we can work together to understand it, experiment with it, critique it, leverage it, and make decisions about if and how we want to use it.
Upcoming Ways to Connect With Me
Speaking, Facilitation, and Consultation
If you want to learn more about my work with schools and nonprofits, take a look at my website and reach out for a conversation. I’d love to hear about what you’re working on.
In-Person Events
February 25. Kawai Lai and I will be facilitating a three-hour workshop, “Human First, AI Ready” at the NAIS annual conference in Seattle, WA, USA. This workshop is designed for school leaders who are navigating the complexities of AI integration at school, including defining ethical behavior, navigating diverse perspectives, and supporting a strategic and sustainable approach.
June 16-18. I’ll be facilitating a three-day AI program called “Learning and Leading in the Age of AI.” This intensive residential program is designed for school teams to have time and space to design classroom-based and schoolwide AI applications for the next school year. Hosted in partnership with the California Teacher Development Collaborative (CATDC) at the Midland School in Los Olivos, CA, USA.
June 23-26. I’ll be joining the Summer AI Institute at Lakefield College School (Lakefield, Ontario, Canada) as a speaker and coach. This event is for teams of educators from Canadian independent schools to advance their AI work, design classroom and schoolwide AI initiatives, and learn from each other’s work.
Links!
I recommend this recent Brookings report, which offers a comprehensive overview of the state of research and practice in AI in education.
Leon Furze has made his “Teaching AI Ethics” series, which I recommend constantly, a free ebook.
If you are concerned about critical thinking and what it should look like in an AI age, you should be familiar with Sam Wineburg’s concept of “critical ignoring.”
English teacher Chanea Bond on why her classroom remains resolutely analog.
Clay Shirky on how AI is helping students avoid vulnerability.
On corporate responsibility: We should be paying attention to the impact of AI content moderation on the many human beings who do it. There are other ways to do this work.
And: An AI toy company left thousands of interactions children had with their stuffed animal exposed to the public.
I had a great conversation with Dan Kearney on his “What’s the Big Idea?” podcast. We talked about what I’m seeing on the ground at schools and what I’m learning from both students and educators.
And, I chatted with Andrew Simon about critical thinking and AI for Polygence’s “Art of Thinking” podcast.
Thanks for the reference, Eric. I'm curious if you've played around with Skills in Claude - much more powerful than GPTs because they can be activated across any Project or Chat and more than one at a time. Very useful for repeatable and discrete prompt sets you want applied across tasks. And the more I play with Claude Cowork the more I understand the potential. I do think 2026 is going to the year of useful and direct AI agentic applications and, unfortunately, most educators are not going to be able to take advantage of it.
In a Feb. 9, 2026, post on his Learning on Purpose Substack, educator Eric Hudson offers a transparent look at his current use of generative AI in education. He outlines specific priorities focused on prioritizing human interaction over automation while emphasizing the need for ethical, critical approaches to AI in the classroom. https://thespikevolleyball.com/