Day 3

18 weeks is a really long time for us to have something to start testing with users.

In the project plan that CofounderGPT prepared during Day 2, the estimated time was 18 weeks. Today we’ll need to find a way to ship something faster. we’ll need to figure out the Minimum Viable Product (MVP). And since we want to move fast, the first version of our product will probably suck really badly.

These are some words of wisdom we’ll live by:

we’ll definitely be embarrassed Reid, we promise.

The plan right now is to spend about 16 hours each per week on this project. So the day number on our posts may not always reflect the actual day from which we started. And we’ll mostly post updates about 2-3 times per week depending on how many hours we put in each week. Our goal for now is to treat this as a side-hustle, our “day job” is at Vacation Tracker. We assume that most people who want to start a side-hustle would have 16 hours per week to spend on it. Our startup story will hopefully give you some ideas and tools to try starting your own business with the help of AI tools.

MVP

When you think about it, CofounderGPT was very ambitious when we started asking it to come up with product ideas. The knowledge base product that we decided on during Day 2 was actually a FEATURE of the second product it suggested. We just decided to zero in on that one aspect of the customer service automation platform that it mapped out. But as Slobodan said during the technical review, even this is too big. We need to find something smaller that we can test out.

Let’s see what ChatGPT can come up with:

Note: It says AIuto in the screenshot because I’ve been confusing CofounderGPT by pasting earlier prompts and forgetting to change the old name

TL:DR by CofounderGPT: The Knowlo MVP focuses on creating a streamlined, functional platform for automatically generating knowledge bases by utilizing core features and pre-existing models, with the primary goal of rapidly entering the market, obtaining user feedback, and validating the central idea.

With this specification in hand, we are still very far away from a product. We are even far away from being able to explain this product in detail to a programmer. CofounderGPT came up with an ok plan, but we’ll need to expand on each one of the parts. And this is really where it comes in handy to know how to talk to AI tools like CofounderGPT. we’re still learning but we’re getting better.

So, Let’s write a very detailed prompt asking CofounderGPT to expand on the Data Ingestion Module (Simplified):

After asking CofounderGPT in a clear way how to build this one module, it provided very detailed step by step instructions which included code for the technical team to implement. At this point, it helps to have someone technical to review this while CofounderGPT is writing code to ask questions or to ask for improvements to parts of the code. You have to keep refining what the AI is writing until you get to something that you can work with and build quickly. I show a more detailed and less technical example of how we refine this below.

For now, the limitation is that you need to take the code, put it somewhere and test it to see if it works. I hear that Plugins for ChatGPT solve that problem because there is a Plugin that will allow CofounderGPT to write the code, run it and fix bugs. In other words, it will be capable of delivering fully functional code. But Plugins are still not available to everyone. I only got access to Plugins today, and only to the very buggy browser Plugin. In a few years, I’m sure AI will be able to spit out fully functional products. For now though, you have to put the LEGO blocks together yourself and You’ll need the help of a technical person to get the product up and running.

After making our Data Ingestion Module (Simplified) specification and code as good as we can make it, we moved on and did the same for all the other modules. Now we have detailed specifications with code to give to a freelance developer who will work with Slobodan on this.

User Interface

Next, we asked CofounderGPT what we should do about the UI. Should we hire a freelancer designer or use a template? Given that we want to move fast and that we’re a scrappy startup, it suggested we go with a template.

But CofounderGPT doesn’t give you the full recipe right away. Well, maybe for a soup it does. Not when you’re trying to build a sophisticated technology product. We had to ask for a detailed description of the criteria necessary for the template and where CofounderGPT suggested we get the template.

Then we had to describe how the navigation will work in our application:

CofounderGPT will need to expand on all of these points. Two sentences to explain each element is not enough. We need something we can easily pass to a programmer and which will provide a clear outline of what needs to be done.

Let me give an example of how I described to CofounderGPT to expand on the homepage:

As you can see, I’m giving providing instructions to CofounderGPT of things it needs to take into account while writing this up. Even after it gave me the result of this prompt, I had to refine further. But this prompt made us go from a 2 sentence description of the homepage, to this:

After the homepage, we did the same for all the other pages of the MVP and put those in Confluence(a documentation tool we use at Vacation Tracker). However, if we didn’t have Confluence, there are a bunch of other tools to store all this stuff that CofounderGPT is writing and keep it organized (for example Notion). No matter what we choose, we need a way to provide an organized project to our technical team so they can put the product together.

Additionally, with the navigation, we’ll also need to write up the user flows for our MVP. How will users navigate through our product? What does the signup look like, explained in detail? What does the sing in look like? How do we onboard users? We won’t paste every question and answer that CofounderGPT provides, these articles would be too long. But we used the same methodology as for the navigation: we provided precise input instructions and to kept expanding on points until we were clear on what needs to be done and we have detailed instructions on how to do it. After that, we move to the next module, navigation element or user flow and we rinse and repeat.

Do we need anything else?

We have our modules defined, we have our navigation defined, and we have our user interface and user flows defined. CofounderGPT, do we need anything else to get this up and running?

Technical review and scoreboard

Slobodan took a look at our MVP specifications and signed off on the scope. We agreed that this is small enough for us to try it out. We even may have an idea for who would be good to work with us as a freelancer on this project. Stay tuned.

For now, here’s the scoreboard for Day 3:

Time spent today: 6h

Total time spent: 19h

Investment today: $0

Total investment: $207

Users: 0

Revenue: $0

What’s next

Even if we manage to build a great product from the start (which we won’t), we’ll need users to try our product and tell us why it sucks so we can make it better. Knowlo won’t find users for itself, we’ll need to find a way to get some people crazy enough to try the embarrassing product we’re gonna launch in a few weeks. we’ll need to think about a go-to-market plan for our MVP. we’ll also need to look for a freelance programmer to help us put all this together. Stay tuned!