We didn't want to rush out a new video about the Bubble pricing announcement on April 6th, so we've waited until the new week, and we think that one of the best ways of looking into workload units is to use an example of a side project I've been working on. And yeah, so let's dive into it.
First of all, this is a side project. It's basically just something that I wanted to build to experiment with OpenAI when we first got GPT 3.5 out about a few weeks ago.
And so this is how it works. I can search for a book, so I can search for Harry Potter, and I can click on the book title. And now it's making a request to OpenAI for OpenAI to recommend three books that I would like because they are similar to Harry Potter and the Philosopher Stone. Once I've demonstrated the app, I'm going to go and demonstrate, well, display to you each of the steps that is taken in the workflow, and then we'll look at the workflow, workload units that are used. Now, I've not read Percy Jackson, but I've seen the film. I think that these are all very similar.
Yeah, so OpenAI has done a good job here. How did they do it? Well, if I dive into the Bobble app and just go back to normal zoom, we can see here, this is today, and the workload usage metrics here are taking a little bit of time to update. I really hope that they fix that because otherwise, how on earth are we going to know how much a particular feature or sub feature in that app takes?
Examining an OpenAI API call
Anyway, this is one that I did earlier, very similar call. So we have 204 workload units. And so what makes up that 204 number? Well, this is what happens when a user clicks on a book. I make the API call to OpenAI. I supply the title and the author. I then have a method here called user stats. This has enabled me to track whether the same user uses the app more than once, and I make changes to that. I then set the slug for the results page where I display my three books. I then create a new book. And there are a number of different ways I could have done this, but I've just approached this as using Split By to effectively parse the data into this is my first book recommended, this is this is my second book, and this is my third book.
I then save those books back into what I've called Saved Search, which is a data type in Bubble. I send an event notification to LogSnag. I'm going to have a video coming out about LogSnag soon. It's a very simple way of tracking user events in your app. And then I send the user to the book recommendation page. So remember, all of that takes 204 workload units.
So I've got a quick calculator here. So if it's single call to OpenAI, with all of those additional elements into the workflow, such as tracking the user, saving the results, passing the results into three new entries in my book database. 204, there we go.
What does this mean in Workload Units
If I have the starter plan, then I'm entitled to 175,000 workload units. So by doing a little bit of very simple formula work in this Google Sheet, that's how many times a user can try and discover a new book using my service a month, leaving me with 28. And remember, there will be additional ways that workload units will be accumulated, such as if a user refreshes the page, such as if they were to try and initiate the OpenAI call, but something was to fail.
It's all going to add up. But I'm looking at about 28 uses a day. And I'm not quite sure what to think of that on the startup plan. Bearing in mind that I wouldn't be able to launch the app on my own domain unless I was on the startup plan. Maybe 28 per day is okay. Like I said, this is just an app that I built out of my own curiosity. I'm not really looking to take it mainstream or anything. But yeah, 28 units. Now, where this... Sorry, not 28 units, 28 times it could run per day.
Now, where this gets interesting is in the course of building this app, before we knew about workflow units, I built this about a month or two ago, I would have been testing it. And I had to test this so much in order to be able to parse the data for individual books to get that formatting from OpenAI to enable me to take a plain text expression from OpenAI and turn that into three separate books. So if I was on the free plan when I was testing it, I'm taking this right the way down to 25,000.
And then, yes, I'm not going to work on this every single day of the month, all 30 days or so, but I'm left with this, what feels like a pitifully small number of basically being able to run testing my very simple app that uses OpenAI. I'm limited to four times a day. And I think this just begins to highlight what I think is a reflective and genuine response from the Bubble app developer community that once the initial shock and the anger and maybe the fear for people's livelihoods has gone, we're left with some plans and some measurements of capacity, the workload unit, which simply does not work. It's just simply so limiting or it's going to cause people's monthly fees to just explode and become huge numbers.
Our reflection on Bubble pricing April 2023
I really hope that Bubble sorts us out. As the time of recording this is, it was April 10th, so a few days have passed since the announcement. The forum posts around are blowing up. There does seem to be some obviously very angry people, but also people who are just concerned about their livelihoods. And I've flip flopped from being optimistic to pessimistic and back and forth about these new pricing announcements.
And I think ultimately they're out of my hands. And I have to try and extend the goodwill to the Bubble team that they've accumulated over the years. And to... It sounds a little bit cheesy, but to trust that they really are what they say they're doing, which is they've read through these forum posts, the hundreds and hundreds of comments based on pricing, all the discussion that's taking place. And I'm trying to be optimistic about an announcement which they said will come later in this week about ensuring that users are not priced out of building the apps that they are already running or are already developing. So there we have it. I've tried not to be too much on the attack or too negative, but I thought it just helped to run through a simple side project that I built. This novel idea, book recommendation using OpenAI, show you the steps and explain why I think they have to change something about the workload units, how they're measured, and simply the quantity that you'll get with each of the packages. Leave a comment below if you have any video suggestions for us. We read every single one.