Lately I've been using ChatGPT prompts and function calling to simulate my aspirational best self. I spent 15 hours last month having such conversations — more than any other app.
My favorite prompt helps me prioritize my deep work. Here’s the structure:
Help me prioritize the next 30-60 minutes of work. I'd like to identify a manageable chunk of work that aligns with my energy levels and excitement, is ideally continuous with previous deep work tasks, makes progress towards my weekly goals and longer-term goals. It’s 9/7 at 4:30pm. My goals this week, due by Friday 2pm: [bulleted list] My goals this year: [bulleted list]
Conversations like this are especially useful when I’m sleepy or distracted, like is often the case at 4:30pm. So I make it as easy as possible to initiate by saving this as a text snippet on my iPhone.
Here’s how ChatGPT responded:
Given your weekly goals and the current time, it sounds like you've got a tight window to make meaningful progress. Since it's 4:30 PM, you might be in that afternoon lull or feeling a second wind. How's your energy level right now?If you're feeling energized: […] This likely requires creativity and enthusiasm, which can be hard to muster when you're low on energy.
If you're feeling moderate energy: […]
If you're feeling low energy: get to inbox zero. This task is more about grinding through emails and less about creative or strategic thinking.
Since your deadline for the weekly goals is tomorrow at noon, maybe save […] for after that.
What do you think? Would any of these tasks feel particularly rewarding to tackle right now?
One of the ways I get stuck is by expecting myself to finish something before an unrealistic and unexamined deadline, so suggestions like “maybe save […] for after” is actually kinda helpful.
I responded with a non-sequitur. Conversations with ChatGPT don’t always need to flow naturally. Sometimes sending a half-baked thought is all my sleepy brain can muster:
Village retreat: making a list of people that would give good talks. Innovation fellows: dunno, I should spend 5m figuring that out. Inbox zero: I should, but I feel so aversive to all those old emails and texts.
In this case, I realized that I have no idea how to proceed for the second task, and that was the source of some resistance. ChatGPT is always happy to provide commentary, which is useful because I can react to it and riff on it:
Ah, the nuances make it clearer. Let's break it down: Village Retreat: If you're excited about curating a list of speakers, this could be a great task for high energy and creativity. Plus, it's a concrete task that you can make good progress on in 30-60 minutes. Innovation Fellows: If you're not sure what this involves, maybe allocate the first 5 minutes of your deep work time to clarify the tasks. Inbox Zero: If you're feeling aversive, maybe this isn't the task for today. But consider setting aside time tomorrow morning when you're fresh. Given your current feelings, I'd lean towards starting with the Village Retreat task. How does that sound to you? Would that feel rewarding right now?
These are good suggestions! At this point I’ve reduced the ambiguity of two tasks, and scoped them down to 25 minutes and 5 minutes of work that I feel motivated to do next. I commit to a plan by telling ChatGPT:
All right, let's identify ideal speakers and talks for the next 25m. I'll check back in at 5pm.
It’s charmingly encouraging.
Sounds like a solid plan! […] could be both exciting and impactful. Have a productive 25 minutes, and I'll be here when you check back in at 5 PM. Go for it! 🚀
By 5pm I’d successfully gotten out of my slump and into a flow.
Like being managed by my best self
Many of us spend large parts of our weeks working at home and solo. Holding yourself accountable is challenging because it’s really two jobs: execution and management. Doing hard work requires immersing yourself in a problem, which involves forgetting everything that isn’t the task at hand. Management, on the other hand, involves being mindful the long-term goals and periodically checking whether you’re on track or whether you should abandon a particular approach.
If I can trust ChatGPT to be mindful of my long-term priorities instead, it’s easier to lose myself in my tasks. It’s like being managed by my best self — not quite there yet, but already useful enough for me to want to spend much of my deep work talking to it. The goal is to fully suspend the background process that’s worried about whether I’m working on the right thing.
This is the real secret of life — to be completely engaged with what you are doing in the here and now. And instead of calling it work, realize it is play. — Alan Watts
Uber-style turn-by-turn instructions?
I wonder: how much of my life can I offload to a manager that represents my best self? Can I also entrust my best self to suggest when I might work out, call an old friend, or maintain inbox zero?
This would require knowledge of my close friends and when I’d last reached out, the status of my inboxes and how long they’d take to drain, and how recently I’ve worked out.
That data is scattered across other apps. So I’ve been collecting much of my life’s data into a single personal Postgres database and implementing APIs to and from all the apps I use regularly. So far I’ve incorporated APIs from Airtable, AWS, Coda, ConvertKit, Discord, GCal, Gmail, GDrive, iCloud, LinkedIn, LogSeq, Notion, OpenAI, RescueTime, Stripe, Texts, Twilio, and Twitter.
In theory, all it takes is interpolating this data into prompts like the above so that my best self AI has enough context to advise me on my next actions, given a description of all my priorities. I’ve recently started experimenting with this but don’t have much to report just yet. I’m cautiously optimistic, though, especially as AIs get smarter and context lengths get longer.
I wonder: would it be possible to have Uber-style turn-by-turn instructions for much of my day, such that I was generally immersed in the moment, making efficient progress on purposeful work, maintaining my healthy habits, and living a life of freedom and serendipity rather than rigid discipline? I’m interested to try.
Lots more to say, especially about other kinds of pre-programmed conversation formats that encode some kind of best practice, but I’ll leave it here for now.