Field Notes
The learning is infinite but the memory is limited. Thats why I write to share my learnings over time to time.
The Double-Lock Principle
How we balance startup speed with enterprise safety
The most critical lesson I’ve learned is that intelligence does not need identity.
Early on, I realized that treating AI as a database is a security failure; if it holds secrets, it can leak them. I learned to separate the "Who" from the "What" using a strict "Double-Lock" architecture.
Keep the users's identity locked in a secure "Vault" (our database) , while placing a "Wall" (PII filter) before the data ever reaches the AI. The model—"The Brain"—receives only anonymized logic puzzles to solve. The breakthrough for me was realizing that the AI can solve the problem perfectly without ever knowing who the pieces belong to.
Why I Build With AI
What it actually means to architect without writing code
Building with AI is less like engineering and more like managing a brilliant but junior assistant. It has infinite knowledge but zero context. It is simultaneously the smartest and most 'stupid' entity in the room—capable of writing perfect code that solves the wrong problem.
"Ask me questions" is a good way of helping the AI create a better result that needs less trimming afterwards. This way of 'spoon-feed' it helpt building within the enterprise reality: Compliance, user behavior, and operations. Don't treat AI as a magic 8 ball that gives the right answer directly.
Treat it as a wild horse that needs to be tamed and shown the fences of the field.
