Notion Solving My Context Window Problems in My AI-CEO Project
Long way here. Happy to be here.
Hey digital adventurers! Quick update on my AI-CEO project - lots of changes, lots of learning. If you’ve been following my AI-orchestrated e-commerce experiment, you know I’m building a system that can run businesses autonomously. And after hours of testing... it’s actually working. Getting closer every day.
Let me share what’s changed in changelog format:
The Big Moves
Google Docs → Notion Migration
Started with Google Docs for AI context (like I showed in my dynamic Claude chat guide)
Works... but not great for complex knowledge systems
Switched entirely to Notion - game changer
Why? Notion is basically a database with nice UI. Now using my Limitless Pendant + Zapier to add thoughts directly into categorized databases
When I ask Notion AI for something specific to a project? It actually finds the right stuff instead of searching through walls of text
The Coding Reality Check
Tried building prototypes in Replit and V0
V0 made gorgeous mockups but... that’s where it ended
V0 - much cheaper and at least I like it Result? I apologized to Cursor and came back
Main lesson: “vibe coding” is fine for frontend, but complex backend needs actual developer work. AI helps, but you need to understand what you’re building
The Gemini Surprise
Late night experiment with Google AI Studio and Gemini 3 Pro Preview
First model that created actually useful code in BOTH frontend and backend
Impressed how little context it needed to produce something that “kinda works”
Might use Gemini 3 Pro as main LLM in the project... that’s a big shift from my usual Claude approach
ChatGPT 5 Mini Economics
Using it in Zapier to categorize my Pendant thoughts
It’s ridiculously cheap for what it does
Problem: struggling with huge context windows. Probably need an extra step to summarize before categorization
Integration Headaches
2.5 years ago in my original AI co-CEO experiment? I used Google Sheets + daily batch processing. Simple, worked great
This time? Going for direct API integrations with real-time data
It’s way more complex... but necessary for true autonomous operations
Every tool integration is its own mini-project. Fun times
What’s Actually Working
The system has its own database storing everything it learns during workflow execution. Current flow that works:
Setup Phase:
Choose business type, investment, target audience, product type
Gemini 3 pro preview - result looks FINE, but what’s worth - it is working AI analyzes and produces 3 best options
User picks one
Build Phase:
AI creates plan with tasks split between AI and human
User accepts plan
AI starts working immediately on its tasks
Everything gets added to Knowledge Base
Human tasks are interactive with input options
When build tasks done → big green button appears
Operation Phase:
Switch from Build to AI-CEO mode
AI operates business... currently in simulation
That last part is key - AI-CEO not fully operational yet. Most complex piece because it needs integration with every business system before doing anything useful. Working on it.
The Honest Truth
Different AI models for different jobs - Claude for reasoning, Gemini for generation, ChatGPT 5 Mini for cheap scale operations. Stop being loyal to one model, start being smart about what works.
Context management is everything. Integration is always harder than you think (even when you remember it’s hard).
And most importantly? It’s actually working. Not perfectly, not completely, but genuinely working in ways that seemed impossible a year ago.
More updates coming as I push through getting AI-CEO fully operational. The good, the bad, the 2 AM debugging sessions... all of it.
What do you think? Anyone else experimenting with multi-model AI systems? Drop a comment.
PS. How do you rate today’s email? Leave a comment or “❤️” if you liked the article - I always value your comments and insights, and it also gives me a better position in the Substack network.