I’ve got a little secret to share with you.

July was kind of a “slacking off” month for me at our side-hustle Knowlo. It wasn’t so much that I wasn’t working. It’s more that I was not able to put any time into the side hustle.

I was travelling with my family in Serbia throughout July, Vacation Tracker did an awesome team building in Divčibare Serbia, and we even managed to squeeze in an 8 day vacation to Croatia. But that just meant less hours for our side project and no time to write anything. The focus in July was on family and the main hustle.

But this week I’m back and we’re planning on accelerating the pace at Knowlo over the next couple of weeks.

Since this is Day 30, Slobodan and I thought this was a good opportunity to assess how far we’ve come and what we’ve learned over the last 29 days of building Knowlo with CofounderGPT.

Invested Time

When we started this little experiment called Knowlo, our goal was to use AI as much as possible to do most of the work creating the startup. One of our ten commandments from Day 1 was that we would each spend about 8 hours per week working on this side hustle.

Well, as it turned out, AI helped a bit less than we had hoped, and we actually spent much less than 8 hours per week each on Knowlo.

As of Slobodan’s last post on Day 29, we spent a combined 194 hours in the first “30 days” of working on Knowlo. In reality, if we assume that 1 working day is 8 hours of my time and 8 hours of Slobodan’s time, and if we would have sat for 194 hours straight instead of in chunks whenever we had time, it would mean we’ve spent about 12 working days building Knowlo over the last 4 months.

we’ve tried to organize the articles we’ve been writing based on the things we were working on, which is why it seems we’ve spent more time working on Knowlo than we actually did.

Nevertheless, in the 194 hours we’ve invested into Knowlo, we’ve managed to get quite a few things done.

Results

With this limited time investment:

1. We came up with a name for our startup: Knowlo.
2. We set up our website using Webflow
3. We set-up an email address, Mailchimp, Google Analytics and Google Search Console
4. Brainstormed product ideas and decided on AI tooltips
5. Wrote and re-wrote the specifications for the MVP
6. Defined and chose a product UI template (Tailwind)
7. Created a proof of concept using OpenAI embeddings
8. Migrate our website from Webflow to WordPress
9. Created some images and basic content for our website
10. Did some market research on possible competitors
11. Developed and expanded on a go-to-market strategy
12. Found a freelancer to help us with the product (and then Slobodan and CofounderGPT ended up building the product)
13. Found a freelancer to help us with the website
14. Wireframed what Knowlo is going to look like
15. Converted the mockups into a front-end theme
16. Developed a back-end architecture for the project
17. Found some pilot customers in our network
18. Started our AI content series and CofounderGPT wrote 9 articles for our blog since we started
19. Created a simple landing page for our homepage and an About Us page
20. We tried to engage with the Indie Hackers, Betalist, Product Hunt Discussions and Microconf communities about Knowlo
21. Did keyword research for our ads campaign with the help of CofounderGPT
22. Wrote front-end and back-end code with the help of CofounderGPT
23. Designed and modelled the database for Knowlo
24. Started sending out newsletters (and then stopped in July)
25. Fixed email deliverability issues
26. Wrote Google Ads copy with the help of CofounderGPT and ran a Google Ads campaign for 4 weeks
27. Wrote Facebook Ads copy with the help of CofounderGPT and Facebook Ads campaign for 2 weeks
28. Set-up remarketing for Google ads
29. Set-up social media accounts for Knowlo and wrote intro posts
30. Fixed Google Search Console issues
31. Set-up sign-up and login for Knowlo
32. Connected the UI to the Knowlo application
33. Setting up Tooltip creation and finalizing our MVP

TLDR: This is a long list but to summarize, CofounderGPT and I did a bunch of prep work to put together a website and build our waitlist for Knowlo. While we were doing that, Slobodan and CofounderGPT have been chugging along with the MVP and are close to having something we can start testing in Vacation Tracker and with other companies.

Budget

Most of the work done so far didn’t require a large budget, especially since we tried to have CofounderGPT do most of it. And whenever that wasn’t possible (which was more often than we expected), Slobodan and I were there to step in to get things done.

Over the course of the first 30 days, the total amount we spent up to today was $1,351 USD. And that includes the extra costs associated with the mess that happened with Webflow, and our migration to WordPress. We wasted at least 15 hours and about $200 USD unnecessarily, but that’s part of startup life. We took a wrong turn and we had to course correct.

While we’re talking about the budget, something interesting happened while I was on vacation. OpenAI refunded us for all the payments we made for ChatGPT Plus from March to July. Basically, all the money we paid for ChatGPT Plus has been refunded to us. Why? I have no idea. It came with no explanation. When I got back from vacation, I saw this in my inbox:

I’ve been providing a lot of feedback to ChatGPT since we started using it in March. This is the only possible reason I can think of why OpenAI might refund us. That or it’s a bug in their billing system. Anyway, we got a total of $100 back from OpenAI which will reduce our total investment to $1,251 USD at the end of today.

We committed to invest $25,000 into this project and we’ve only spent a tiny fraction of that so far. This is great news because we have most of our budget left to promote Knowlo and to acquire customers with it.

Is CofounderGPT a good partner?

The main thing we’ve learned is that CofounderGPT has been a great co-founder so far, but It’s not quite on par with a human yet.

When we started, I made the bold prediction that CofounderGPT can be at least as good as a human to build a startup with. I still think that it will be one day, hopefully in the not too distant future. But at the moment, you still need a lot of human work along with the work that AI produces to get your startup off the ground.

To be clear, having CofounderGPT helping us drastically accelerated all of our activities from coming up with a name, to writing code for our MVP. Everything we got done with the help of CofounderGPT was way faster than if it was just Slobodan and I working on Knowlo alone.

Things we’ve found that CofounderGPT is good or pretty good at:

• Ideation and brainstorming
• Writing, editing and suggesting improvements for content (everything from articles and specs to website content and ad copy)
• Writing, reviewing and commenting code
• Doing research and analysis

I’m sure there are other things CofounderGPT will be able to help us after we launch but so far, the most bang for our buck we got out of our AI cofounder was with these four areas.

What we didn’t expect

To put it bluntly, we didn’t expect CofounderGPT to get dumber and lazier over time. When we started working with GPT-4 and decided to launch Knowlo with an AI co-founder, it felt like something magical was happening as we worked with CofounderGPT (based on GPT-4). But then suddenly the performance got worse.

For a while, it was so bad that it was almost unusable. It seems to be getting better now but I still don’t feel like It’s at the level it was in March. I’m sure this is temporary and that GPT-4 will get much better over the next couple of months. But that doesn’t change the fact that right now, It’s not as good as the first version we worked with.

This is part of the reason we’ve been going a bit slower. When ChatGPT-3.5 was released in November, it was an incredible demonstration of current AI capabilities. And then a few months later, OpenAI rolled out ChatGPT-4 which was orders of magnitude better than 3.5. This created a sense that this is how quickly we should expect things to evolve with AI. But it hasn’t been moving that quickly and the reality is, best case, we are still 6-12 months away from having something that is ready for mass adoption in businesses.

The other thing we didn’t expect was how busy things were going to be with Vacation Tracker. This is the second part of the reason we’ve been moving more slowly than planned.

So when are we launching?

Good question. Slobodan and CofounderGPT have mostly wrapped up development of our MVP. What we need to do over the next 2 weeks is tie a few loose ends in the code and then test it out internally. we’re hoping to start letting people in in the next 2 weeks.

Stay tuned and signup to our waitlist to be among the first to be notified of our launch!

Scoreboard

Most of today was spent on pre-launch activities, and the -$100 was for the refund of our ChatGPT Plus subscription payments.

Time spent today: 4h

Total time spent: 198h

Investment today: -$100 USD (ChatGPT Plus refund)

Total investment: $1,231.45 USD

Beta list subscribers: 92

Paying customers: 0

Revenue: $0

What’s Next?

Slobodan and CofounderGPT will be finalizing the MVP for testing and we’ll be going through our pre-launch checklist to get everything ready for the millions to start rolling in.