AI - Practical

I watched a video today:

I’ve really been leaning into a lot of AI recently. Even beyond the idea of using AI to write code for me, I’m really starting to use AI to help the actual running of the code.

One of the projects that I’m working on, a mix of a side project as well as an application for work, is a system to go from a HAR file to a Postman collection that is runnable. One part is kind of easy in principle: figuring out what are common things from a response that is replicated in a subsequent request.

Some parts are a lot harder.

The easy side of hard is coming up with variable names. Sure, you can do things like call a variable “var1” and “var2.” But those suck.

The first application of AI is going from the javascript paths from which I retrieved the value to names of variables.

The harder part is going from the physical path that a value was found, to the logical path that needs to be used. One example that I’ve found is a ticket number in one response has a sibling value of “ticket”. Determining the JSON path expression for that — and even knowing that it should be done — is a hard problem. But it’s surprisingly easy with an API call. The magic happens. But I don’t really need to know how it happens.

And it’s still cheap. An entire run still is around one milli-dollar. I.e. $0.001 dollars.