Pip is an AI-native product built around one number before you spend.
Pip turns a misleading bank-balance habit into Spendable Cash Today: a daily spending signal backed by deterministic application logic and an agent that explains the why when the user asks.
The bank balance is real. It is just not all open room.
Most people already have a money habit: open the bank app, look at the balance, and treat that number like permission. Pip keeps the habit small but changes the default number.
The product is intentionally narrow. One daily number comes first. The agent sits behind it for explanation, setup, and actions like account connection or data refresh.
The agent does not own the financial truth.
That boundary is the important systems lesson. The model can explain, route, and ask for clarification. Deterministic tools calculate Spendable Cash Today and perform approved side effects.
Deterministic money engine
Application code owns the calculation and returns structured facts for the agent to explain.
Agent action layer
The OpenAI Agents SDK route handles explanation, setup, account actions, and typed UI cards.
Real data boundary
Supabase handles auth and user data. Plaid handles read-only account connection. Pip cannot move money.
Trust copy baked in
The public product language avoids guarantee claims and repeats the read-only, no-money-movement boundary.
Production deployment
Netlify hosts the product and marketing site with environment checks and deploy proof tooling.
Verification discipline
The repo includes unit, E2E, live-smoke, deployment, bundle, and PRD completion gates.
Pip is a product and a systems case study.
For employers and clients, the signal is not only that Pip exists. The signal is the judgment underneath it: reduce scope, protect the system boundary, build the useful version, then wire it into real services without letting the AI invent truth.
One daily signal instead of a dashboard.
The agent is part of the interaction model.
Supabase, Plaid, OpenAI, Netlify, and typed app code.
Read-only data, no money movement, no AI-owned math.
Need this kind of AI product judgment on a messy workflow?
Send the context. I can help find the real decision, define the system boundary, and build the useful version first.
Email Tyler