25 Years of Helping People Use Powerful Tools Before They Become Easy
Every technology wave has a gap. The tool exists, but most people can't use it yet. KayGee exists to close that gap. Again.
It's Not AI, It's a Flavour of AI
"AI" was never one thing. What the word means keeps changing flavour, from search to rules to generation, and each one behaves nothing like the last. Marek has built with all of them.
A* search. Game development.
The first hands-on AI. Pathfinding you could read line by line. Legible by design.
Rule-based AI. A chatbot.
Hand-coded rules that never came together. Legible, but it didn't work.
Rule-based AI. Logistics and finance.
The same approach in bounded domains, and it worked really well. Legible and reliable where the rules were clear.
LLMs.
Powerful, but the reasoning is no longer legible the way rules were. Marek works with them to find their real potential and business application.
Shoot the gun, but keep your goggles and ear muffs on.
Action is necessary. Learning happens by doing. The tool is powerful but not benign. Misuse compounds.
You don't need to fully understand the internals. You need to not hurt yourself.
Teaching judgment under AI conditions.
The Discovery
Marek was building a website with AI. Mid-project, everything broke. The AI said: "File too large. I can't process this anymore."
That's when it clicked. Every AI session resets. You build something, it works great. Next session, the AI has forgotten everything. You explain it again. And again. Your effort doesn't accumulate. It just resets.
The problem isn't you. It's that AI has no persistent memory. Unless you give it one.
What Marek Tried First (So You Don't Have To)
Custom instructions. Better than nothing. But they don't grow with your project. Static context in a dynamic world.
Missing Habit 1 (plan in markdown) & Habit 4 (measure compound)
Prompt libraries. Great for one-off tasks. Useless for multi-session projects where decisions compound.
Missing Habit 2 (protect decisions) & Habit 3 (align with files)
Long conversations. Context window fills up. AI forgets the beginning. Important decisions buried in thousands of messages.
Missing all 4 habits. Nothing persists between sessions
The answer is simpler than you'd expect. Put your context in files that live in Git.
Not in chat history. Not in custom instructions. In versioned markdown files that AI reads every session. Decisions persist. Patterns accumulate. Context compounds.
What Makes You Think You Can Build?
So you build it. Which raises a fair question. This is the third time Marek has closed the gap between a powerful new tool and the people who can't use it yet.
Built CMS Before WordPress Existed
Publishing content to the web required hiring developers. Marek built Janmedia, a CMS that let non-technical people publish directly. The barrier was knowledge locked behind code. The solution was to democratise publishing.
Two years later, WordPress launched and solved this at scale. The pattern was validated.
Taught WordPress Workshops
WordPress existed, but most businesses couldn't use it effectively. They needed more than documentation. They needed guided practice. Marek ran in-person workshops teaching teams how to maintain their own websites instead of waiting for agencies.
The pattern repeated. The tool was free. The knowledge transfer was valuable.
Architecting AI Collaboration Protocols
AI tools exist (ChatGPT, Cursor, Claude). Everyone can prompt. But every session starts from scratch. Context resets, decisions are forgotten, effort doesn't accumulate.
The pattern continues. The protocol is free (open source on GitHub). The compound effect is what makes it valuable. And that takes guided practice to learn.
Every wave, the same mission. Stop people from waiting for gatekeepers. Give them the knowledge to use powerful tools themselves.
What Changes After Session 3
The first two sessions feel like overhead. You're writing files instead of building. Then session 3 happens.
Your AI references a decision you wrote last week without being asked. You didn't prompt it. It read the file.
You change one component. Your AI flags five others that depend on it. Not because it's smart, but because you documented the architecture.
You open a new session. Instead of "what are we building?", your AI picks up exactly where you left off. That's 30 minutes of setup paying back every single day.
The setup is 30 minutes. The compound effect is everything after.
The Origin
1987. Marek arrived in Australia from Poland at age 13, speaking no English and without a dictionary.
Knowledge was locked. Communication was impossible. Every conversation required translation, patience, and someone willing to help.
That experience shaped everything. When knowledge is locked, progress stops. When knowledge is accessible, everything changes.
Why This Mission Matters
Technology moves faster than institutions can teach it. By the time universities offer courses, the tools have already evolved. By the time agencies understand it, the next wave has started.
The gap is inevitable. Someone has to close it. KayGee exists to do that work. Not just once, but for every wave that matters.
What the Protocol Has Built
SpellMe and a handful of others were built with the protocol, from zero. The gym came first, and the protocol was extracted from it.
SpellMe
A spelling app that prevents learning damage by never letting children practice wrong patterns. Coach/Learner model with gentle correction, spaced repetition, and 8-language support.
Protocol used duo/ab
Gym Workout Tracker
Personal fitness companion. Structured workouts, exercise swapping, effort tracking. The app the duo/ab protocol was extracted from, 400+ commits over six months, deployed and used daily. There was no protocol when it started.
The Compound Effect in Action
These products weren't built by a team. They were built by one person with one AI collaborator, using four habits that compound effort over time. Plan in markdown. Protect decisions. Align with files. Measure compound.
The gym taught the habits. Everything after was built on them. That's what the compound effect looks like.
Want to Try It?
The protocol is open source. Start learning at your own pace, or get in touch to share feedback or ask a question.