Startup Sprint: Racing Through Day 2 with CofounderGPT
Day 2: Let’s get organized
When you’re starting a startup, there’s no shortage of things to do. Luckily we have CoFounderGPT which can help us get through a lot of this stuff quickly. One of the things I learned a long time ago when I started my first company is that you can spend a lot of time doing things which contribute nothing or very little to you actually moving forward. And since both Slobodan and I are time constrained with this startup, we’re going to try prioritizing the ~20% that gets ~80% of results (a.k.a. the 80-20 rule or the Pareto Optimum).
Here’s the list for today, and how we’re gonna divide up the work:
- Get a domain for Knowlo (Lav)
- Signup for a Webflow account (Lav)
- Choose and buy a Webflow template for the product landing page and blog (Slobodan and Lav)
- Explore different product features for Knowlo and decide on which features we want to start with (CofounderGPT, Slobodan and Lav)
- Write a project overview document and plan timelines (Mostly CofounderGPT with a little help from Lav)
- Write detailed specifications for the product which will include as much code as possible (Mostly CofounderGPT with a little help from Lav)
- Technical and feasibility review of all documentation (Slobodan)
- Write Day 2 blog post (Lav)
- Write a CofounderGPT introductory blog post (CofounderGPT)
- Write Day 2 social media summaries (CofounderGPT and Lav)
Setting up a website
The first three items on the list can be roughly grouped as ”setting up the website”. We bought the domain from AWS when we decided on the name on Day 1. We already host everything for our other companies on AWS and plan on hosting Knowlo as well.
Although we could ask CofounderGPT to write the code for the website, we would still have work to do to clean up the code and get it up on a server which requires time. We also have access to a development team which could easily set this up for us, but there is a cost associated to that and we’re a scrappy startup. So we decided to go with Webflow because when you break it down, there is no way that either of the two options I mentioned before would cost less than the $28 per month we’re gonna pay to get started with Webflow + the $129 for the site template.
And there is another reason we decided to go with Webflow while we validate the idea: speed. I signed up for a Webflow account, downloaded the template and set up a simple landing page with a blog in 5 hours.
I actually expected to spend less than 5 hours, Webflow has gotten more complicated. Not to mention, their pricing is weird and they’re forcing us to pay double of what we expected ($56 per month) in order to be able to publish our site onto our own domain. I only found this out when I tried to publish. Yes, It’s there on their pricing page in fine print, my bad. Read the fine print folks! Well played Webflow, well played.
For anyone interested, if we were not in such a rush to get this product launched, we would move somewhere else now. But we’ll stick with Webflow for a bit and then we’ll move somewhere else. If we were starting over right now, we’d probably just go with a WordPress.com hosted website for now. So for anyone trying this at home, save yourself the headache (and cost) and go with WordPress.
We put the logo together in Canva in about 5 minutes. It’s nothing fancy, just a font that says KNOWLO. Good enough for now, we’ll make a fancier one later if necessary.
There’s minimal content on the site and any content you see was written by CofounderGPT with my guidance. It’s not the nicest looking landing page or blog. Actually, It’s kinda shitty to be honest. But done is better than perfect.
We need to start capturing email addresses as soon as possible and we need a home for these blog posts. We set-up a free Mailchimp account to store the emails and connected it to Webflow. The goal is to have a small audience of people to email as soon as we have something testable. we’ll put up some screenshots and other product related content on this website once we have a clearer idea of what we’re building.
Exploring product features
CofounderGPT provided a few possible variations of the knowledge-base-type product but ironically, we ended up liking the first recommendation best because it was kind of what we initially wanted to do with CofounderGPTs twist on it. You can see what that was in our Day 1 post.
TL:DR: CofounderGPT sums it up as developing an AI engine that generates a comprehensive knowledge base by analyzing customer interactions, support tickets, and product documentation, and dynamically optimizes the content based on usage patterns and feedback. Provide personalized recommendations to improve user experience, ensure multi-channel accessibility, and offer analytics and reporting to help businesses make data-driven decisions for enhancing support operations and customer satisfaction.
Slobodan, CofounderGPT and I agreed, Let’s roll with this for now. And Let’s ask CofounderGPT to write up a detailed project overview:
Sometimes CofounderGPT forgets its role in the game and goes back to being ChatGPT-4. To get things back on track, you have to give it a little reminder. It’s kind of frustrating and this is the only annoying thing so far with having an AI as a cofounder.
The project outline that CofounderGPT prepared looks ok, but if we are going to actually build this into a product, we need to go a lot deeper to outline how things are going to work. We also need some specifications that we can give to a freelance programmer to put together the code that CofounderGPT will write for us.
This is really where you get to see the big difference between a human cofounder and an AI cofounder. If I had to sum it up, human cofounders are better at understanding the initial inputs and producing better initial outputs. they’re also better at getting clarifications on things that they don’t understand by asking questions and they’re more friendly and sociable. CofounderGPT is not as great as a human at these things. But it works orders of magnitude faster than a human. And if you know how to talk to it, you can get pretty comparable outputs to a human at a fraction of the time it would take a human to do it. Last week, I estimate CofounderGPT did about 3 weeks of work in 2 days (at a cost of about $2). So I’ll let you draw your own conclusions about what you would prefer in your startup.
After a few attempts asking CofounderGPT to write a high level specification, this is what it came up with as the main modules of the product:
It also suggested some user flows but Let’s ignore those for now and write detailed specifications for each module. We need to get CofounderGPT to write all the code and then write specifications for Slobodan and a developer to put it together. I’m not gonna post a bunch of screenshots here with long specification documents, that’s not the point of this story. The lesson here is that you have to keep pushing CofounderGPT to go deeper into the rabbit hole if you want to get the really good results.
Here is the prompt I started with and then I refined from there:
To be clear, CofounderGPT is not magic and it just spits out a complete project which you upload and it’s running in the cloud. You do need to have some technical knowledge to keep nudging it to provide more and more details. Eventually, You’ll end up with a big part of the project done by CofounderGPT. But Slobodan and our future freelance developer will be the ones who finish the project and get everything up and running.
I kept going with CofounderGPT until we had detailed specifications with code and instructions for the technical team. There is a lot to be reviewed! Before going any further, we’ll get Slobodan to have a look and provide some comments.
Introduction to CofounderGPT
While we wait for Slobodan, we figured it might be a good idea to get our star CofounderGPT to introduce itself through a little blog post, now that we have a blog. we’ve been chatting for a while and it has a good sense of what we’re doing and has been in character. Here is the post CofounderGPT wrote as an introduction.
Technical review and scoreboard
Slobodan had some comments about the specifications that CofounderGPT and I wrote, the main of which was that the scope of the project is too big and that 18 weeks is a long time to develop a prototype. We need to move faster to test this idea out. And that leads us to What’s next, but first, the scoreboard:
Time spent today: 8h
Total time spent: 13h
Investment today: $187 USD
Total investment: $207 USD
Refining the product idea even further into an MVP. For non-technical people, here is an explanation from the OG Eric Ries who came up with the whole Lean Startup Methodology: What Is an MVP? Eric Ries Explains – Lean Startup Co. . Basically, we’re going to figure out a way to shrink the scope of the project. In other words, we’re going reduce the features to a bare minimum so that we can launch fast and start iterating.