Recently I came across a fascinating use case: can you create a synthetic loan using a box trade strategy?
For anyone not deep in options: a box spread is a classic arbitrage construct. By combining calls and puts at different strikes, you can replicate the cash flows of a loan. In other words, you can “borrow” synthetically in the options market.
That sounded like the perfect testbed for a quick build. So I decided to build one.
The Build
I spun up a Box Trade Agent (aka Synthetic Loan Agent) using Loveable, an AI-native dev tool.
Total build time: under 2 hours.
But that doesn’t mean the problem was trivial.
The app connects to my E*TRADE production account via OAuth (E*TRADE’s sandbox APIs were too limited). The user enters a loan amount and term, and the agent pulls the options chain, evaluates ~20,000 spread combinations, and selects the set that yields the lowest implied borrowing rate — ideally converging toward the 1-year Treasury yield aka the risk-free rate.
Where GenAI Fit In
There’s no GenAI inside the app, nor for that matter is there any AI/ML. The “intelligence” is just financial structure and code.
But it was built with GenAI.
That’s the shift: tools like Loveable compress the SDLC - significantly. They give you wild, unfair speed. And that speed raises the bar. But it doesn’t replace judgment, systems intuition, or technical depth. If anything, it makes them more essential.
As a PM, you need hard skills. Without them, you’re just scaling mistakes faster.
The Hard Skills I Needed
Even with Loveable doing all the scaffolding, I had to:
Work through OAuth and secrets to authenticate with E*TRADE.
Bypass sandbox limitations and safely test against production APIs.
Understand the financial domain well enough to know that modeling just 20 strikes wasn’t sufficient — I had to broaden the search to get realistic borrowing rates.
Guide the tool through rigorous testing, edge cases, and validations until the agent was robust and correct.
Why This Matters
This project wasn’t about finance alone. It was about how we build products in the GenAI era.
Build speed is now a given.
Product sense and technical depth are the differentiators.
PMs who can debug, test, and reason in code will thrive.
The future of product management isn’t just about roadmaps and specs. It’s about building alongside AI, fast loops, and having the hard skills to ensure quality and robustness.
Over to You
What would you build with two hours, an agent framework, and access to real APIs?
