The Interesting Problem Wasn't the Chatbot
When I began revisiting 觀音 for the age of generative AI, the obvious question was:
How do I add AI to this application?
It did not take long for me to realize that this was the wrong question.
Connecting an application to a large language model is relatively easy. Send a prompt, receive a response, display it in a chat interface.
That may demonstrate AI integration.
It does not necessarily create a better product.
The more interesting question was what people were actually trying to accomplish when they asked a fortune stick about a relationship, a career, money, family or an uncertain future.
They were rarely looking for information alone.
Often they were trying to make sense of something.
That changed how I thought about the product.
Two Different Roles
I did not want AI to become an electronic fortune teller.
There was little value in asking a language model to manufacture another prediction.
Instead, I began thinking about the traditional fortune stick and AI as two systems with different responsibilities.
The fortune stick brings intuition, symbolism and insight.
AI brings reasoning, objectivity, emotional steadiness and practical next steps.
The distinction became central to 觀音AI .
A fortune stick can interrupt the way someone is thinking about a problem. Its ambiguity can reveal another perspective.
AI can then help the person examine that perspective.
What do we actually know?
What are we assuming?
What alternatives have we ignored?
What can we do next?
The purpose is not to replace human judgment.
It is to help create the conditions for better judgment.
Then Privacy Changed the Architecture
Once the product became conversational, another question became unavoidable.
What would people actually be willing to tell it?
The conversations where this kind of AI might be most valuable can also be deeply personal.
Relationships.
Fear.
Career uncertainty.
Regret.
Things people may not want attached to an account, transmitted to a server or stored somewhere they cannot see.
That made privacy different from the usual product requirement.
It was not something to add later as a policy page.
It affected whether people could use the product honestly in the first place.
And if privacy was part of the product promise, architecture had to support that promise.
That led me toward one of the defining technical decisions behind 觀音AI : running AI directly on the user's device.
Why This Is the First Issue
I chose 觀音AI for the first issue of MankitSze Magazine because it represents the kind of technology problem I find most interesting.
The story crosses product design, software engineering, AI architecture, privacy, user experience and an application with roots going back more than fifteen years.
It also illustrates something I believe is increasingly important as generative AI becomes commonplace:
Adding AI is not the same as finding the right role for AI.
The difficult work often happens before choosing the model or writing the integration.
What problem are we actually solving?
What should AI be responsible for?
What should it deliberately not do?
What architectural decisions follow from those answers?
And what does the user gain that was impossible—or impractical—before?
Those are the questions I want this magazine to explore.
Not AI as a feature checklist.
AI as a product and engineering decision.
This first issue begins with 觀音AI .
But the larger subject is how we decide what AI should become once we have the ability to put it almost anywhere.
— Mankit Sze
End of Editorial