Automating my Sunday finance review
I've been doing a weekly family budget review for years.
For a long time, it took 1-2 hours every Sunday. Now, with the help of AI agents, it takes 5 minutes.
I like knowing where my money is going, but maybe budgeting isn't your thing. That's fine. The point isn't the budgeting. It's finding the recurring tasks in your life or work that matter to you but keep getting skipped, and asking whether an AI agent could do the heavy lifting.
---
🔄 The whole thing runs automatically
It runs on a schedule, connects to Monarch (my financial tracking app) via API, pulls all of our transactions, does its work, and delivers me a short review queue. I don't touch it until it's ready for me.
📏 Deterministic rules where you can, AI where you can't
Not everything needs AI. For merchants I can reliably categorize every time, I built deterministic rules in Monarch. They fire automatically, perfectly, every time. For everything else, an agent matches the raw transaction description to a rule in a persistent household rulebook. Its job is narrow on purpose. Keeping the scope small makes it more accurate.
🛒 Computer use for retailers like Amazon
For Amazon, you can't apply a deterministic rule. You can buy anything there. So the agent opens my authenticated browser, navigates to my order history, reads the actual line items, and splits the transaction into properly categorized pieces. If the numbers don't reconcile, it flags it instead of guessing.
🔍 A second agent audits the first
A separate independent agent checks the work with no access to the first agent's reasoning. I get the settled result after they've finished disagreeing with each other.
---
The agent flags only what it genuinely can't resolve. I open Monarch's review queue, handle a handful of real judgment calls, and I'm done.
1-2 hours down to 5 minutes.
Here's what I'd take from this:
1️⃣ Use deterministic rules where you can. Reserve AI for the unstructured stuff that requires judgment.
2️⃣ Give AI guardrails. A rulebook, a narrow job, a defined scope. It shines when matching messy inputs, like a transaction description on a CC statement, to clear rules, not inventing answers from scratch.
3️⃣ AI can be deeply personal in a way other solutions can't. A financial advisor doesn't manage your grocery bill. AI can hold that context and learn your preferences over time. There really isn't another good solution to this job, other than not doing it 🤷♂️
4️⃣ I'm not a software engineer. I designed this using first principles and AI to fill the gaps. You don't need to be an engineer to build something like this.
Every Monday I try share how I'm using AI in my work and personal life. Now hopefully you know how to use AI to manage your budget....and what my wife buys on Amazon 😂
#agentMondays