Day 3 of AI - growthbot
Okay, i think I need to start breaking the information down differently. I’ll organize by bots for now.
adsbot
adsbot, the tool meant to help me track spend versus sales, is still stuck waiting for Amazon to approve permissions for me to scrape the ads data. It is noted that sometimes this takes weeks.
igbot
Yesterday, I noted that I didn’t have a scheduler, so I added that. I can now create the posts I want in one sitting, and queue them up. Claude says that MWF, posts in the morning, work best for Instagram fantasy writers, so I went with that. I can override and push a post at any time. Y’all, the setup is pretty cool. The posts are tracked in a database on Amazon, and categorized by series (my urban fantasy versus epic fantasy) and type of post.
We stumbled on the fact that carousel ads are “in” right now, and the igbot cannot handle them, so we added that to the “to do” list. I don’t know where that “to do” list lives… I suspect Claude is like me and creates a list that it never refers to until it’s wife reminds him of it. 😁
We made an attempt at cross-posting to FB from IG. We will see whether that works this morning when the next scheduled post goes out. If it doesn’t, we have another way.
We added Pinterest posts to the bot. Here’s a little technical bit—Pinterest doesn’t want you developing person-specific apps, so we couldn’t build out the way we had for IG/FB. We had to use a tool called “make” that already has Pinterest approval. “make” also has its own AI tools, but we are using just a simple WebHook. It looks like I get a limited number of token, and then I have to start paying. I cannot tell whether that is tokens per month or total or what, so still learning.
I noted that our post tracking only monitors the IG posts, so I plan to add Pinterest and FB tracking. Claude pushed back and said, “Let’s wait ‘til the first cross-platform ad goes out this morning and be sure that all works before we make any changes.” Probably wise.
growthbot
For my next foray into AI, I asked Claude to help me design something that will look at what I’m doing, and suggest ways to improve, with the goal of finding and building my audience.
Yeah, that was pretty vague, right?
Claude proposed two agentic prompts (two AI prompts). One would take the data collected from the adsbot and the igbot (the data management of likes and such) and look for optimization opportunities. The second would be an outside-the-box thinker that would look for much broader opportunities. So, (1) optimize what exists and (2) look for new ways to grow.
We stitched that together this morning, and it required a few new bits:
Created a slack channel that the growthbot reports to. (It also sends me an email, but I like the idea of managing history in slack). If you haven’t used slack, it is similar to Discord, where you create channels for conversations.
Stitched in Kit, my newsletter email service, and FB Ads Manager (already mostly stitched in for the data collection of IG “likes”). This will look at my newsletter grown and suggest improvements. I pushed on this because FB and Kit often disagree on their statistics. FB tends to make decisions that don’t quite line up with reality b/c it’s numbers aren’t quite right. I can explain why, but that’s not for here.
It was interesting, because it was in the Claude conversation for growthbot that I realized we weren’t tracking stats for anything but the IG posts, and when I hopped over to the igbot convo, it said, “Great idea, but let’s wait til the synced posts go out today before we dink with it.”
I think we are likely pretty far from a working model for growthbot. We don’t have the LwA piece to collect ads data. We don’t know that the data we want will be collected yet. We likely have a lot of iterations to go. But it will be interesting to see what these AI systems say about the data (I hope).
LIMITATIONS
Okay, I’m starting to see where I am scraping against the limitations.
I hit the usage limits yesterday. It was in a session where I was iterating over an interface I didn’t understand, so I was sending lots of screenshots. I think that used up my “bandwidth” quickly, as I really hadn’t done a lot before it boxed me out.
Code sharing across sessions isn’t great. You have to stash the code in the project files. Since I have multiple sessions (adsbot, igbot, growthbot), each one of them has a different repository, and trying to keep the data up to date is cumbersome. One solution might be to make a “super-project” and super-repository that shares all the code.
The pieces of the different bots don’t align. That makes the last issue worse. Trying to align them requires MY having to sync all the data. Again, the solution might be to create a master project and essentially “start over”, feeding the new project all the bits we were working on.
