I started using Claude to help with financial management. I’ve found it very successful.
There’s a lot of stuff that Claude will not do. this includes:
- No financial advice.
- No logging into websites, or creating accounts.
- No transfers or purchases.
- No signing documents for you, or submitting them for you.
- No entering of sensitive information or filling out anti money laundering docs.
It is also very careful with private documents.
However, despite this, it is very useful. It will do:
- download every statement from a provider’s website.
- compare two sets of numbers, for example:
- do pension contributions on payslips match the provider’s numbers?
- Have we claimed every medical expense?
- do financial analysis - how different savings options compare, top outgoings, comparing months and so on.
- find different products - for example given power usage numbers and times from home assistant, it will find the best power provider.
Claude often feels like glue - it is coupling together stuff that should work by itself but doesn’t. For example, most providers should provide a way to download every statement in one go, or export a CSV file.
Claude pushes me to do stuff that I am reluctant to do, like ring my healthcare company and ask if I can claim an old thing. I can’t help but think this is good long term.
I don’t know if anyone else will use this. Claude code is really intended for writing code, and I sense not totally loved by Atrophic. OpenClaw is similar in idea. I’ve tried out a couple of other headless chrome AI setups, but they hide the browser itself, so only the AI and interact with it.
I wonder where agents will end up. There are businesses where a system that does more business is good for the business. For example, a stock broker charging a per transaction fee, or a restaurant. There are also businesses that depend on customers being lazy, for example power or insurance companies. These are going to have a bad time.