Human in loop Experiments

Kim Wong

— 1 min read

I’ve been experimenting with a different way to use AI in products: not just as a chat bubble, but as the thing that drives how the UI responds. 💬📊

As a testbed, I built an AI-first expense tracker with a simple layout:

• Left: chat where you talk in plain language 💬

• Right: live dashboard the agent drives 📊

Under the hood it already handles:

• showing and filtering transactions in tables 🧾

• account and category browsing 💼

• bar charts or different analytic views 📈

• adding / updating / deleting via natural language (including batch recategorisation like “change all Uber to Transport”) 🔁

The interesting part for me isn’t “chatbot for money”, it’s the interface pattern:

• the agent interprets intent, chooses which tools to call, and decides which widgets to show

• every write goes through an explicit changeset review — no silent edits to your data

• you talk in natural language, but the response is a structured, visual UI instead of a wall of text

Personal finance is just the sandbox. The bigger idea is:

if an agent can orchestrate both the operations and the UI, you’re no longer stuck choosing between rigid forms or pure conversation.

Just sharing because it’s been fun seeing this run as a real interface instead of yet another lonely chat box in the corner. 😄

play it with this link: https://ai-expense-tracker-chat-frontend.vercel.app/

github repo: https://github.com/kimwwk/ai-expense-tracker-chat-frontend

#AI #AgenticSystems #InterfaceDesign #ConversationalAI #ProductDevelopment

Read more