P06_personal
P06_personal
P06_personal
Building a shared wallet with AI
Building a shared wallet with AI
Building a shared wallet with AI
Ji started from a personal need: I wanted a simple way to manage shared money as a couple, while exploring whether an AI coding agent could help me build a usable product end to end. I designed it around a shared-fund model: both users add money to one balance, shared expenses are deducted from it, and every transaction remains traceable.
For the first MVP, I focused on making this core flow reliable before adding more features. I used Supabase for shared data and Codex throughout planning, implementation, debugging, and validation. AI helped me build faster, while product rules, scope, and data reliability remained my decisions.
Key decisions / trade-offs
Shared fund vs. bill splitting
I chose a single shared balance instead of tracking who owes whom, keeping the product focused on managing shared money rather than settling debt.Editable records vs. balance accuracy
Transactions needed to remain editable, but every change also had to keep the shared balance consistent. This made data integrity part of the core product logic.Core reliability vs. feature expansion
I kept the first MVP focused on deposits, expenses, balance updates, and transaction history before adding broader budgeting features.AI assistance vs. product control
Codex supported implementation, debugging, and edge-case review, while product rules, scope, and validation remained human-led.
Takeaway: AI agents were most valuable when they accelerated the path from product rules to implementation and validation without replacing product judgment.
Ji started from a personal need: I wanted a simple way to manage shared money as a couple, while exploring whether an AI coding agent could help me build a usable product end to end. I designed it around a shared-fund model: both users add money to one balance, shared expenses are deducted from it, and every transaction remains traceable.
For the first MVP, I focused on making this core flow reliable before adding more features. I used Supabase for shared data and Codex throughout planning, implementation, debugging, and validation. AI helped me build faster, while product rules, scope, and data reliability remained my decisions.
Key decisions / trade-offs
Shared fund vs. bill splitting
I chose a single shared balance instead of tracking who owes whom, keeping the product focused on managing shared money rather than settling debt.Editable records vs. balance accuracy
Transactions needed to remain editable, but every change also had to keep the shared balance consistent. This made data integrity part of the core product logic.Core reliability vs. feature expansion
I kept the first MVP focused on deposits, expenses, balance updates, and transaction history before adding broader budgeting features.AI assistance vs. product control
Codex supported implementation, debugging, and edge-case review, while product rules, scope, and validation remained human-led.
Takeaway: AI agents were most valuable when they accelerated the path from product rules to implementation and validation without replacing product judgment.
Ji started from a personal need: I wanted a simple way to manage shared money as a couple, while exploring whether an AI coding agent could help me build a usable product end to end. I designed it around a shared-fund model: both users add money to one balance, shared expenses are deducted from it, and every transaction remains traceable.
For the first MVP, I focused on making this core flow reliable before adding more features. I used Supabase for shared data and Codex throughout planning, implementation, debugging, and validation. AI helped me build faster, while product rules, scope, and data reliability remained my decisions.
Key decisions / trade-offs
Shared fund vs. bill splitting
I chose a single shared balance instead of tracking who owes whom, keeping the product focused on managing shared money rather than settling debt.Editable records vs. balance accuracy
Transactions needed to remain editable, but every change also had to keep the shared balance consistent. This made data integrity part of the core product logic.Core reliability vs. feature expansion
I kept the first MVP focused on deposits, expenses, balance updates, and transaction history before adding broader budgeting features.AI assistance vs. product control
Codex supported implementation, debugging, and edge-case review, while product rules, scope, and validation remained human-led.
Takeaway: AI agents were most valuable when they accelerated the path from product rules to implementation and validation without replacing product judgment.