How is my AI use changing?
A look back at the evolution since I started running AI workshops in March
What is something I am doing less than I was in March?
Using NotebookLM. ChatGPT has significantly upgraded its capabilities in ways that have reduced the areas where NotebookLM has an edge. For example, with Connectors for Google Drive, ChatGPT now lets me find sources and ask follow-ups across my files. This is just one of several ways that ChatGPT has eroded my go-to use cases for NotebookLM.
What is a feature you are surprised hasn’t been added to ChatGPT yet?
Google calendar integration. I got access to Claude’s Google calendar integration in mid-April, and I thought ChatGPT would fast follow, but it hasn’t. (You can integrate your calendar for Deep Research but not for everyday tasks like finding time blocks in your schedule.)
I don’t think this is explained by the fact that Microsoft is OpenAI’s biggest investor. Microsoft gets paid back its ~$13B through profit-sharing, and there’s more money to be made by playing nice and being useful to Google Workspace users.
What is a feature you are surprised Claude hasn’t added yet?
Honestly, nothing. Claude doesn’t have Persistent Memory across chats, can’t talk out loud, and has laughable image generating capabilities. But these ‘missing’ features feel aligned with Anthropic’s (parent co) strategy. Claude is intentionally:
Stateless: Unlike ChatGPT, Claude has no cross-chat memory so that it can’t quietly build a user profile or reinforce biases (but therefore also doesn’t maintain useful context about your work or preferred interaction styles or anything else).
Text-first: Claude accepts almost twice the context of ChatGPT and prioritizes the quality of its writing/coding output, but it has no voice mode and limited image generation capacities.
Suite-agnostic: Anthropic isn’t entangled with Microsoft (like OpenAI) or Google (like Gemini), and I’ve seen no evidence that they seem to be building their own workbench. If this holds, it means Claude will be designed to integrate well.
This list may make you wonder why I haven’t made Claude my default LLM. Beyond the very real switching costs, the answer is largely memory: I get so much out of it and know how to manage its downsides, which makes it hard to consider making Claude my default.
What is a feature you still feel like you’re under-utilizing?
ChatGPT’s integration with Python open up a whole universe of tasks, from chart creation to image manipulation, even if you aren’t a programmer. But because I do not have a background in Python, I’ve had to stumble on these uses rather than having a ready library of ideas I could flexibly and immediately draw upon. Contrary to the ‘AI is killing coding jobs so learning to code doesn’t matter any more’ sentiment, I believe that having the first ~20% of coding knowledge has never been more valuable. You can use it to creatively yet realistically assign tasks to AI and to make more efficient use of vibe coding products.
What is a feature you think others are still under-utilizing?
Projects are not just folders to store chats but a place to create mini-operating systems and assistants. When I teach people to fully leverage Projects, I start hearing things like “I realized I was using only 1% of what ChatGPT has to offer.”
What is a dream feature for ChatGPT to add?
So many, but if I had to pick one realistic yet dreamy upgrade, it would be live sync between Projects and Google files with the ability to write directly to files from the chat. I want to be able to say ‘Add this block of text to that doc’ not just generate a new doc.
A little more complex example: I’ve built a wine ‘sommelier’ that knows what I like, and it works great as a Project except that I can’t easily store information in a structured way. I don’t want to vibe code an app for this task, especially since a spreadsheet would suffice. A live sync would allow me to write to a spreadsheet without having to open the spreadsheet, copy over information, and then re-update the project file.
(Gemini can one-click create a Doc from a chat but can’t live-sync edits or write to Sheets. Claude has Projects and you can use Google files as project knowledge, but you can’t write directly to the files from the chat.)
What has changed the most since your first workshop series?
From a content perspective, the session that has changed the most is the Knowledge Partner session. As I mentioned above, ChatGPT has a lot more capabilities in this area than when I launched the series in March. So what was a session entirely focused on NotebookLM is now maybe 20% on NotebookLM.
From a process perspective, I’ve gotten smarter about how I ask participants to apply what they’re learning between sessions. I used to make time for breakouts during class, but most people preferred more examples over in-the-moment practice. Now, I assign homework with each session and ask everyone to email me their work in advance of class, including snags they hit or what worked particularly well. I start class by sharing a full list of what people worked on & giving an opportunity for questions. I also pull lessons from across the group to share with everyone. This approach leads to more follow-through, richer discussions, and a broader range of use cases.
What have you learned from You & AI participants?
The place I learn the most is seeing and hearing about participants’ wide variety of applications. Sometimes they’re ones I’ve been wondering about, and sometimes they are totally novel. Some recent ones:
Working with Persistent Memory to create profiles to separate out two different users of the same ChatGPT account
Creating a Scheduled Task that provides a weekly synthesis of important emails (in Copilot)
Using Deep Research to find restaurants that meet a particular set of criteria (in this case, family-friendly Michelin-starred restaurants along a vacation route)
Are you launching any new workshops?
Yes! This summer I’m working with young adults for the first time. It feels great to actively work with people this age, rather than just think about how they are using AI & how it is affecting them. I’d like to find more opportunities to do this type of work over the next year, so please think of me if you have ideas or opportunities.
What is something you’d like to launch?
An async version of the course: Not everyone can make Friday sessions, which is what I mostly offer. Organizations often want to have a way for people who joined the team after live sessions to catch up. And some people would happily trade off the opportunity to ask questions for the opportunity to hit pause, rewatch segments, and do it on their own time. But it also feels like a daunting task, especially since AI changes fast enough that it feels like the videos would need to be frequently re-recorded. Feedback on this idea is welcome—or if you have a different idea for me, let me know!
What is your website?
When is your next training kicking off?
August 1st at noon ET. There are four spots left. You can learn more & enroll here, where you can also see additional workshop dates posted through September. I’m truly grateful to everyone who helps spread the word. Thank you!
Will you write a newsletter like this one again?
Maybe! Let me know if you like the self-interview retrospective format. :-)
Coauthored index: Sarah 80% | AI 20%
