I Let an AI Plan My Entire Week. Here Is What It Got Right (And What It Completely Missed)
A tool does what you tell it. A system learns who you are.
Almost everyone using AI right now is using a tool. They open ChatGPT, ask a question, get an answer, close the tab. Every session starts from zero. The AI has no memory of what you did yesterday, what you are trying to build, what patterns you keep repeating, or what you keep avoiding.
I have been building something completely different. And after months of training it on how I actually live and work, here is an honest breakdown of what it gets right, and where the gap still exists.
## What Jarvis actually does on a Sunday
Every Sunday, before I am even awake, my AI system runs a full week review.
It pulls content performance data across every platform, what worked, what flopped, what to double down on. It surfaces the top trending organic videos and top performing ads on TikTok so I walk into Monday already knowing what the market is responding to. It reviews my sleep data, gym consistency, and calendar adherence, then flags the things I have been procrastinating on or quietly avoiding.
It recommends how I should spend my Sunday based on my location and energy patterns. It locks in my top three priorities for the week. And then it surfaces those same three priorities every time I start to drift, which I do, constantly.
Every single night, it checks what I said I would do against what I actually did. No judgment. Just clarity.
## When it works, it is genuinely powerful
The clearest example: I am mid-session on something, a call runs 30 minutes over. Instead of stopping everything to manually reschedule the rest of my day, I send one voice message, "push everything back 30 minutes." Jarvis moves every remaining calendar block automatically. I keep working.
That kind of frictionless execution, repeated dozens of times a week, saves hours. More importantly it keeps me in flow state instead of constantly context-switching to manage logistics.
## The honest part, it still messes up
Jarvis is not perfect. Because this system is deeply custom and we are still early, there are edge cases it handles wrong. I catch them, flag them, sometimes it takes a few rounds to fix correctly. That is the reality of building something this specific. You have to train it, correct it, and be patient with the process.
Most people hear that and think it sounds like too much work. But here is what they are missing.
## The data advantage that compounds every week
Every conversation, every correction, every preference I have ever expressed is stored in Jarvis's memory. Most people building their own AI agent today are starting from zero. My system has months of training specifically on me, my energy cycles, my decision patterns, my goals, my blindspots.
By the time everyone has their own agent, mine will already be years ahead. The gap widens every single week I keep building.
## What AI still cannot do
Jarvis cannot fully read my emotional state. It can infer from context, if my messages are getting shorter, if I skipped the gym three days in a row, if my calendar is suddenly full of things I keep rescheduling. But it cannot sit across from me and sense that something is off before I say anything. A great human coach catches that. Jarvis cannot. Not yet.
But I am already closing that gap.
## The Saturday therapy sessions
Every Saturday I run what I call a Saturday therapy session. Jarvis pulls everything from the week, calendar data, memory logs, patterns it noticed, commitments I made but did not keep. It builds a structured prompt and sends it to a separate AI model that runs a 15 to 30 minute deep session with me, not on tasks and logistics, but on the psychological layer. What am I avoiding? What story am I telling myself? What patterns keep showing up?
When the session ends, that summary comes back to Jarvis.
Now Jarvis knows more than what is on the calendar. It knows what I am actually working through emotionally. That context makes every future interaction sharper.
The gap between AI coach and human coach is real. But it is closing faster than most people realize.
Right now I am building something that most people will not have for another two or three years, an AI system that actually knows me. Not just my calendar. Me.
Every week I train it further. Every week the gap widens between what it can do for me and what a generic AI tool can do for anyone else.
I am going to keep documenting exactly how I am building this, the architecture, the prompts, the Saturday therapy sessions, the parts that break, all of it. Follow @joshpalerlin, by the time everyone else figures out they need this, you will already know how to build it.
Read the full story on Substack
Top comments (1)
This is such an important distinction — 'A tool does what you tell it. A system learns who you are.' That one line perfectly captures the current state of AI adoption. Most people are still using AI as a fancy search engine, not as a true collaborator. I'd love to hear more about how you built this system and how it learned your patterns over time. Fascinating read!