Getting Started on an AI Marketer Bot
I don’t know what I’m talking about. Let me start with that.
But I’m the kind of guy who figures out his way by starting down a path and then learning and correcting. So you get what you get.
Goal
My first goal is simple: create a bot that captures my data each week for ad spend and sales. I do that every Sunday. It doesn’t take me long. But this felt like a good starting point. I have general hopes that I can do more, like have it make recommendations about other Product Targets (authors similar to me to run ads against), and then even bigger things like other agents that help with social media.
But let’s start with the simple. I use chatGPT a lot, but chatGPT tells me that it doesn’t have the ability to create these somewhat autonomous agents. It gave me some options, and we landed on this plan:
use github to not only store code, but as the master cockpit. I think of this as “the agents live here”. I have a vague sense of the limitations of this approach. I think github will work off of triggers, which I will have to design. So there isn’t sort of an AI bot making independent decisions… I will have to trigger it. That’s not exactly what I was hoping for, but this seemed like a reasonable starting point.
The code will run on Amazon Web Services (AWS). So github stores the code, AWS runs the code.
There’s something called Continuous Integration (CI), which is a concept, not a platform. The idea is that you create tests for the thing you’re building, and the tests always have to pass. It is a way to make sure you don’t check in code that breaks whatever you’re doing.
To my understanding, the first AI that I care about will live on the github side, working inside the CI loop. Once I get enough pieces built, the AI can use CI to iterate and built the thing I want built. It will use CI to be sure it never breaks anything. You give it a goal, and maybe the AI even improves your AI over time to test against the things you told it you want… I’m not sure.
Status
github is set up with a project called ‘adsbot’ that has simple code to write data to a Google Sheets sheet.
Google “keys” are set up to access the Sheet.
AWS is set up to run the code. I originally used AWS CLI to build the code on my laptop (windows), but I’m pretty sure I am now building it either on the AWS side or on the github side. I think it’s happening on the github side, because it needs to build the code to run it’s CI tests.
I have manual triggers set up everywhere. I imagine that needs to move to something more automated.
I use .VSCode on my laptop to see the code that is in github (and to push changes in via git).
Now, for my particular project, I need access to data from Amazon for my books. I need a service called Login with Amazon (LWA). I screwed this up and I accidentally registered to create apps (I don’t know for what) instead of LWA, and now I have to wait for Amazon to resend some data.
Also, note that Amazon only provides the interface (LWA) for looking at ad data. They don’t supply an API for looking at sales data. So I will have to grab that data manually (a button to dump it to excel, then another button to upload that to AWS). Really stupid, right? These are all 3 amazon services that don’t talk to each other. ChatGPT warns me that trying to automate that step could get me in trouble. I’m not sure that’s true, but I won’t poke on it for now.
Next Steps
Obviously I’m waiting on Amazon to give me access to LWA. I could probably poke on other pieces in the meantime, but I’ve got other things to work on, so wiating is fine.
What I’m hoping is that once I have access to the data, then I can turn on the ‘bot’ that will work on improving the code. In other words, I want to see if it can go from the bare bones to what I want to collect on its own (acknowledging that I have to provide it the sales data so that I don’t get in trouble with Amazon).
