We didn’t have as much time to work on Knowlo last week because we are in the middle of our SOC2 audit at Vacation Tracker. After all, Knowlo is our side-hustle and we won’t always be able to put in the full 16 hours. But whenever we are not able to put in the 16 hours, we’ll try to make it up the following week.

After an eventful Day 3, we’ll now be looking at the some AI tools we can use, we’ll be doing some market research and we’ll be developing an MVP go-to-market (GTM) strategy. we’ll also go over Slobodan’s assessment of the product and the requirements for choosing a freelance programmer.

Things are moving fast with AI

Since we started this little experiment called CofounderGPT, we’ve been using ChatGPT-4, but we didn’t have access to the GPT-4 API or the new Plugins that OpenAI announced recently. In the last couple of days, we got access to both. Although for Plugins, we only have the browser plugin which is basically ChatGPT with internet and It’s kind of buggy. I find that Bing is actually a lot better for doing things like market research or for extracting information from a link. But I find Bing doesn’t write as well as ChatGPT, so if it does produce some writing from the research its doing, I usually pass it through ChatGPT-4 to improve the output.

In terms of image generation, DALL-E 2 is great for beginners but I find that Midjourney is the best. Bing also has an image generator built in and It’s probably around DALL-E 2’s level. All the main images for the blog posts and the robot images throughout the site are made with Midjourney. It’s very powerful but not that easy to use. DALL-E 2 and Bing are much easier than Midjourney. So if you’re looking to get started quickly, go with one of those two image generation tools.

There are hundreds of other AI tools coming to the market and I’m sure we’ll explore some of them as we build our business with CofounderGPT.

I’m writing this part about AI because when we say CofounderGPT, we generally mean ChatGPT-4. that’s our primary AI tool for the moment. But it doesn’t mean that we won’t use other tools like Bing for this experiment as part of CofounderGPT in order to get the best results. At Knowlo, we’ll use the best tools available to get the job done.

Market research

Before building the product, we want to take a look at the market to see who’s already doing something like this. I actually hope there are competitors already. It means that a market exists for this product and that we reduce the risk of getting trapped in the First Mover Disadvantage.

Let’s start asking the web-connected CofounderGPT to run a market analysis to see what it comes up with:

This is far from a comprehensive analysis and it only gave one potential competitor which is not even building something comparable to us. As I said, the new Browser Plugin for ChatGPT is pretty wonky and doesn’t do a great job doing live online research.

Let’s ask Bing to see what it can come up with:

I used CofounderGPT to summarize what Knowlo is about and then I put that into Bing to get a list of competitors. Success! Bing does a much better job of searching the web. Probably because Microsoft trained it with all the web data it has indexed in the Bing search engine. I kept asking Bing questions and got a big list of competitors, some of whom are not doing an AI knowledge product like ours…yet.

It turns out there are actually a LOT of competitors for this product. A discouragingly large number in fact. And after doing some research, I have to assume that Microsoft and Google are going to offer these types of products also.

So the good news is, this is a huge market. Otherwise, how could all these companies exist and survive? The bad news is that there are a lot of companies, a big majority of which are much better funded than us. There is another piece of good news though. For most of these companies, $1M in ARR is nothing. Zendesk’s ARR is over $1.5B. For them $1M is a rounding error. Their product is built to be more general purpose so that it can serve many customers across many industries. It’s not practical for them to focus on niche markets.

So this is where our opportunity lies in this large market. But we’ll need to shrink the scope of Knowlo even more and maybe come up with our own twist so that we’re not competing with these guys directly. Let’s set that aside for now and we’ll discuss it with Slobodan tomorrow when we actually go over the details of his technical review.

One way or the other, we need a go-to-market strategy. This is just as important as building the product. There are many shitty products that are very successful because of good GTMs and many incredible products that never made it because of bad GTMs.

So Let’s start coming up with some ideas. We always need to be thinking about how we’re gonna find customers for our product.

Starting our go to market strategy

First, I asked CofounderGPT to tell me what elements the most successful GTM strategies have. I took what I though was most important and wrote this prompt:

What I got back was an average and pretty generic strategy without going into much detail. So I kept refining each section, asking more questions and pushing CofounderGPT to be more creative. Specifically, I wanted a strategy that mostly took advantage of CofounderGPTs skills. The end result was a decent go-to-market strategy which is good enough for now:

we’ll certainly need to make adjustments to it once we finalize what the product will look like. But at least we have a better sense of who we could target, what our distribution channels could be, and what key performance indicators (KPIs) we can use to track our progress:

Once we actually start building the product, we’ll work with CofounderGPT in parallel on each of the elements of the go-to-market strategy.

Let’s keep moving.

CofounderGPTs specs and code

Slobodan is going to be writing the next article in this series where he’ll explain the technical aspect of what CofounderGPT produced and how we’re gonna put it all together. The executive summary that’s relevant for today is that he thinks this could be feasible but that it will take a few days to get it up on a server, start testing it and fixing bugs. Also, he signed off on the technologies that we’ll be using.

Slobodan is working full time as CTO at Vacation Tracker so he doesn’t have time to play with the code that CofounderGPT wrote. we’ll need a freelancer to help us. Luckily for us, we’ve worked with a lot of programmers over the years so we have a big network, especially Slobodan. But Let’s put together a little job description so that we can send this to people we’ll be interviewing to help us.

Here is how I asked CofounderGPT to write me a job description for this role:

This produced a pretty good job description. I asked CofounderGPT to make some small improvements and this is the final result we went with:

For fun, I asked it to also suggest some good interview questions and to provide a simple test to assess whether the person we’re interviewing has the necessary technical knowledge for this project.

we’re now ready to start looking for someone to help us.

Scoreboard

We didn’t spend anything today so the scoreboard is unchanged:

Time spent today: 6h

Total time spent: 25h

Investment today: $0

Total investment: $207

Users: 0

Revenue: $0

What’s Next?

Slobodan is going to write the next article and will provide a better overview of the specifications and code that CofounderGPT produced in Day 3. we’re also going to shrink the product scope even further and see if we can come up with a new twist on this idea.