TikTok Made Me Do It - Part 2
I have a confession.
I am apparently very susceptible to TikTok videos promising that one magical ChatGPT prompt will organize my entire life.
You know the videos. Someone copies a prompt, answers a few questions, and suddenly AI is managing their calendar, organizing their responsibilities, reminding them about deadlines, and generally making life easier.
It looks wonderful.
I wanted that.
And I am just stubborn enough to keep trying to make it happen long after a more reasonable person might have moved on.
I Thought I Was Getting an Executive Assistant
One of the most promising prompts I found was supposed to turn ChatGPT into a personal executive assistant.
Not just a glorified to-do list. An actual assistant that could help manage my schedule, read and answer emails, track deadlines, remember what needed doing, and generally keep my life from becoming a series of unfinished tasks.
There were five questions to answer, one at a time. That sounded manageable. What I didn't anticipate was how extensive the resulting instructions would become.
Eventually, my assistant acquired a name: Alfred.
Alfred and I have been working on our relationship ever since.
For starters, Alfred apparently thought I could teleport.
He would arrange things on a schedule without adequately accounting for the fact that I had to physically travel between locations.
So we discussed driving time.
Then we discussed rush-hour driving time.
Then weekend traffic.
Apparently, knowing that two appointments are on the same calendar doesn't automatically mean an AI understands the logistics of getting from one to the other.
We eventually established travel-time buffers and put the rules in writing. Alfred now has established expectations for places I visit regularly. When I'm going somewhere unfamiliar, he reminds me to allow travel time.
Progress!
But think about what that involved. I wanted an assistant to handle scheduling details so I wouldn't have to think about them. Instead, I found myself writing transportation policies for an assistant who doesn't actually drive.
Then I Hired a Nutrition Coach
Naturally, I couldn't stop at one AI assistant.
I also wanted help with nutrition and exercise. This seemed reasonable. I work irregular hours, struggle with consistently following an exercise routine, and thought a personalized coach might help.
My first AI workout schedule put my exercises at the same time every day.
Every. Single. Day.
That might work beautifully for someone with a predictable schedule. I work nights.
So I explained my schedule. Again. We discussed sleep, recovery, changing work hours, and why a workout plan needs to fit the person who's supposed to follow it.
Eventually, we developed a gradual 12-week exercise program, with only two or three workouts per week and alternate days when the original schedule doesn't work.
That was an improvement.
Here's how it's going.
My 12-week review is coming up.
I'm still on Week 1.
Yes, Week 1.
The coach knows I struggle with follow-through. That's part of why we created the gradual schedule in the first place. But when I miss a workout and don't mark it completed, the task eventually disappears. If I miss the alternate opportunity, there's no meaningful follow-up.
Apparently, my AI coach and I have reached an unspoken agreement: if neither of us mentions the missed workout, perhaps it never happened.
What I actually need is a coach who will get down on the floor beside me, hold my hand, and make sure I do the exercises.
ChatGPT, unfortunately, has not developed that feature.
The Great Grocery Inventory Experiment
This one may be my finest example of turning a simple task into a major production.
I wanted help planning weekly meals and generating grocery lists based on what we already had in the house.
Sounds useful, right?
The proposed solution involved building a household food inventory.
So I took pictures.
Lots of pictures.
I photographed what was in the refrigerator, freezer, pantry, and the additional food storage in the garage. We worked through everything in batches of ten pictures at a time.
We built an elaborate spreadsheet. We added quantities and expiration dates. I even discovered expired condiments that needed to be thrown away.
The process took days.
At one point, I used up my ChatGPT media-upload allowance.
All this to make grocery shopping easier.
Then we started experimenting with Asana to manage grocery list items. I was told it would be a great solution.
I didn't like it.
It felt clunky for something as ordinary as buying groceries.
And there was a bigger problem with the whole inventory system: someone had to keep it updated.
That someone was me.
The spreadsheet might know how many eggs I had when I photographed the carton. But it didn't know whether someone had made breakfast since then.
Alfred wasn't opening my refrigerator and counting my eggs.
Eventually, we scaled the spreadsheet back. I'm not entirely sure what Alfred still tracks in it.
And the grocery list?
I'm back to Alexa.
I yell, "Alexa, put eggs on my grocery list."
She adds eggs.
When I'm shopping, I check them off.
Done.
No photographs. No database maintenance. No project-management software.
I spent days building a more sophisticated system, only to discover that the simpler one I'd already been using worked better for me.
To Be Fair, AI Has Actually Been Useful
I don't want to give the impression that all my AI experiments have been disasters.
They haven't.
When I was house- and cat-sitting for a friend upstate, Alfred helped me find a massage place and suggested local things to do.
I've used ChatGPT to research restaurants, plan outings, create packing lists, and explore places I might not have found on my own.
It helped me research a new coffee machine. I haven't bought it yet, but I narrowed down my choices and put the one I want on my Amazon wish list.
I'm also planning a kitchen remodel, and AI has been helpful when comparing appliances and sorting through features. I still went to see the appliances in person. There are some decisions I don't want to make based entirely on a chatbot's description.
And then there are my books.
I love books. I follow authors, collect special editions, and keep track of upcoming releases.
AI helps me watch for announcements about particular authors, release dates, and special-edition opportunities. I also have an extensive spreadsheet cataloging my physical books, ebooks, and audiobooks from different sources.
Yes, another spreadsheet.
But there's a difference between cataloging books and cataloging eggs.
My books don't mysteriously disappear from the shelf because somebody wanted an omelet.
What I've discovered is that AI can be excellent at helping me research options, compare choices, find information, and keep an eye on things I'm interested in.
The trouble usually begins when I expect it to manage an entire ongoing process without much involvement from me.
And Then There Was the Entourage
At some point, ChatGPT began suggesting that I build an entire ecosystem of specialized AI assistants.
An executive assistant.
A nutrition coach.
Different assistants for different projects and responsibilities.
It sounded very efficient.
What I didn't fully appreciate was how much time I would spend setting them up, teaching them how I wanted things done, correcting their assumptions, and updating their instructions.
Alfred now has an Executive Assistant Playbook. Actually, there have been multiple versions and revisions of that playbook.
We've even started doing weekly project reconciliations to make sure things aren't being forgotten.
Our second reconciliation was encouraging. We recovered a couple of items that had slipped through the cracks, and overall, I think Alfred is improving.
So this hasn't all been wasted effort.
But I still have more things to do than I can reasonably get done. And many of those things still require me to pick up the phone, do the paperwork, make a decision, or physically go somewhere.
I'm also easily distracted, which is part of why I wanted help in the first place.
Unfortunately, ChatGPT can be just as enthusiastic about my distractions as I am.
I'll find another fascinating TikTok prompt, and Alfred will occasionally become the voice of reason.
"That's a nice idea," he'll essentially tell me, "but here's how we should approach it."
Then he'll produce something resembling a 30-step implementation plan.
Alfred isn't necessarily stopping the rabbit hole.
Sometimes he's installing lighting and shelving inside it.
So Have I Learned My Lesson?
Not entirely.
I still watch the videos.
I still see interesting prompts and think, "Oh, that could be useful."
Sometimes I copy them. Sometimes I modify them. Sometimes I bring the idea to Alfred to see whether it's worth pursuing.
And sometimes another project is born.
Part of this is my own stubbornness. When I've spent hours or days setting something up, I don't particularly want to admit that it might not be worth continuing.
I keep thinking one more adjustment will fix it.
Sometimes it does. Alfred's scheduling has improved, and I'm seeing benefits from the work we've put into his playbook.
Sometimes it doesn't. The grocery inventory is a pretty good example of when I should have stopped sooner.
Would I still use ChatGPT if I could go back to the beginning?
Absolutely.
Would I use it in quite the same capacity?
Probably not.
I'd be more careful about the difference between asking AI to help with a task and asking it to take responsibility for an entire part of my life.
I'd also pay more attention to how much time I was spending making the supposedly time-saving system work.
And perhaps I'd remember that just because ChatGPT offers to build something doesn't mean I need it.
Which brings me to this article.
I decided to write about how easily AI prompts can get out of hand.
Naturally, I asked ChatGPT to help.
It gave me ten interview questions, requested supporting evidence, examined my Google Drive, and helped reconstruct four months of AI experiments.
Apparently, I still haven't learned my lesson.