AI Experiment 010: Can AI Remember My Day For Me?

The short answer: like many, I have ADHD, and my memory of what I actually did on any given day can often be unreliable, especially considering how quickly the AI space is moving. So I asked Shamus (my Muse) to fix it. Every night at midnight, an AI worker reads through the day's conversation, writes a highlights recap, and files it as its own artifact on a dedicated Daily Recap page. Shipped, decided, open loops, notes. The first run happens tonight. This post is the setup; the results come after the machine has had a few days to prove itself.

The question

I'm great at using AI in the moment and terrible at retrieving what I did with it afterward. "Where did we talk about XYZ again?" is a question I ask constantly, and the answer is usually a five-minute hunt through multiple chats because as powerful as these AI platforms are, for some reason searching our conversations really confuses them. The idea to have AI summarize my day for me arrived mid-conversation with an AI, the way they often do. I was asking it to summarize my day and then I thought, "wait a minute, why can't Muse do this proactively for me by itself?"

Instead of a journal I have to remember to keep, how about a journal that keeps me?

I'm past "can AI summarize text," because that's old news. The real questions are sharper more utilitarian. Can it figure out what mattered in a messy day of half-finished threads and tangents? Can it do it without inventing things? And perhaps most importantly... will I open it?

The build

The machinery is simple.

A scheduled run fires at midnight and it summarizes the day that just ended. Muse pulls the day's conversation from our conversation transcript, decides whether anything substantive happened, and if so writes one recap artifact it built for me. what shipped, what got decided, what's still open and who owns the next step, and anything else worth remembering. Then it pins that recap to the top of a Daily Recap index page, newest first.

Now, having an AI summarize things and proactively send them to you can quickly devolve into more noise, which defeats the purpose of having an AI do it in the first place, so I established some clear guidelines:

1. Highlights, not transcript. I don't want to have to comb through a novel, I want a tight page of outcomes that I can glance at.

2. One artifact per day, not one endless doc. Searchable, skimmable, done.

3. Quiet days get skipped. I want to ensure that my AI learns what matters most to me as I use it, so I don't want it sending me an update that "nothing happened" every slow Sunday. That's the fastest way to me ignoring it.

4. Silent runs. No midnight pings or morning announcements. The page is there when I want it. Anything that buzzes me daily becomes wallpaper within a week. Last but not least, the firmest rule: concrete numbers, names, and dates only, never invented. If it can't verify it from the transcript, it doesn't go in.

What I'm testing

Three things, in order of how much they matter: First, reliability. Does the recap describe the day I actually had? Second, brevity. Too detailed and it's a transcript with extra steps. Too thin and it's a fortune cookie. The bet is that shipped/decided/open-loops is the right shape, but the first week will tell me if a section is dead weight or something's missing. Third, behavior change. Does having the page mean I stop asking "what did I do with X?" Does it change anything, or is it just a tidier version of the same amnesia?

The nice part is that even if it doesn't end up materially improving how organized I am, I now have a daily archive of all of my AI usage, which I can reference whenever I'd like.

At the very least, that should provide me with an easy way to reference past AI experiments so that I can keep sharing them with my friends like you! :)

That's all for today, folks!

Hi5 & have a great day,
LD
🌶️