AI-Techno Jinni: Just because AI can do it, doesn't mean it should.
Building with AI has made more things possible. HoursKZ, warehouse interfaces and an architectural detour taught me to ask which of them belong in the product.
Read Journal #18TechJinni Journal
Notes from building real products with AI agents.
What worked, what failed, what changed my thinking, and what I am still trying to understand. Written from experience, without pretending the work has produced a finished doctrine.
AI-Techno Jinni: Be careful what you wish for.
A Russian website and a railway illustration taught me to look beyond a convincing AI result and ask whether it fulfilled the wish.
Read Journal #17
AI can generate images. Editing them is harder.
Editing AI-generated images into reusable assets means protecting approved details, layered source files and transparency—from game sprites to product illustrations.
Read Journal #16
One blocked task shouldn't stop the AI workflow.
A blocked task can be a legitimate stopping point without stopping unrelated, authorized work across the AI workflow.
Read Journal #15
AI Can Get the Math Right and Still Get the Method Wrong.
A calculation and its test can agree while sharing the same incorrect interpretation of the source method.
Read Journal #14
The longer an AI project runs, the less it should depend on the conversation.
Long-running AI work becomes easier to resume when current decisions and evidence survive outside the conversation.
Read Journal #13
If the answer cannot change the decision, I probably do not need to ask the question.
Each Discovery question should earn the interruption it creates by reducing uncertainty or changing what happens next.
Read Journal #12
A prerequisite can become the problem.
A legitimate prerequisite can grow into a mechanism whose own dependencies take over the work it was meant to support.
Read Journal #11
If the same operational failure keeps recurring, the system has not really learned.
Recovering from a failure is useful. But if the same failure keeps demanding the same human correction, the workflow may have recorded the lesson without really learning it.
Read Journal #10
Running is not the same as ready.
A system can be running and producing real evidence without yet being ready for UAT, release, production or the next meaningful action.
Read Journal #9
I don’t want more automation. I want less unnecessary human work.
As AI agent workflows become more autonomous, I’m becoming more interested in which human involvement reflects real judgment and which is only operational plumbing.
Read Journal #8
Working with AI has less ego. I’m not sure that’s always an advantage.
One of the things I have come to appreciate about working with AI is how little ceremony there is around changing my mind.
Read Journal #7
A rubber duck that evolves with usage, not time
Repeated use can make an AI assistant more useful before the model changes at all, because less of the working context has to be rebuilt from zero.
Read Journal #6
Ask why you need AI before asking where to put it
Using AI to build a product and building AI into the product are separate decisions; what matters is the improvement it creates.
Read Journal #5
AI stopped being just a coding tool for me
Somewhere between implementation, product decisions, debugging and delivery, “coding assistant” stopped describing how I was actually using AI.
Read Journal #4
When a small internal tool starts becoming a product
The first version solved the problem I could see. Real use revealed the workflow around it.
Read Journal #3
HoursKZ started as a very small problem
I wasn't trying to build workforce software. I wanted to stop tracking employee hours manually in my coffee shop.
Read Journal #2
I started keeping notes on what AI changes in product work
The product changes were easy to record. What changed my mind was harder to keep.
Read Journal #1