The final 10% - Build Log #8

Time for another Build Log. If you're new here, you can read through previous ones here.

Completing the final 10%

This week has been a busy one as I continue to work on the product that I have been building. I am now in the final stages of completing the project, before launch.

I have been working on this project for around two months now, and am getting much closer to a finished V1 that I'm happy with. I always find that the initial phase of a project comes together very quickly - it's easy to get something functioning in a relatively small timeframe. However, I find the final 10-20% takes much longer than initially anticipated - there are always lots of small edge cases of UX improvements to make, to ensure that the product is not only functional but also usable. The market that the product will be competing in is already competitive (despite having a low NPS across the board), so it's important to launch with a very strong UX from the outset - some other markets seem to be more forgiving of a poor UX initially.

Therefore, one of the key areas that I have been focusing on is making sure that the user experience is as smooth and seamless as possible. This has involved a lot of testing and debugging, as well as making improvements to the user interface and overall design. I have also been working on adding new features and functionality to the product, based on feedback from early testers and users.

Doubling down on content in 2023

I've written previously about how we're seeing a rise in creator-led brands (in particular in Build Log 2). At the same time, I feel that there are not that many people in the tech space who are actively sharing the experience and process of building a company in real-time. There are of course threads on Twitter, but these rarely offer a realistic perspective of what the reality is like.

Some of my favourite 'tech-based' creators (links to their YouTube channels): Garry Tan, John Coogan and Hallden. Only the last example actually shares more about the experience in a vlog format. I'd be interested in learning about other creators who do this that I haven't come upon yet.

In 2023, I've got a lot of plans to share more actively what I'm working on: this includes more ideas for expanding on what I share on this website (feedback always welcome too), and launching a YouTube channel to offer a more personal look at my process for building technology businesses going forward.

This year has been a year focused on exploration for me, and next year I plan to double down on the things that are working. I'm excited to document these and share the journey as I go.

Release of ChatGPT, and the productisation of AI

Yesterday I wrote a post about some of the most interesting use cases that I've seen with ChatGPT, a new tool released by OpenAI recently. It's designed to respond conversationally, with an in-built short-term memory so that you can converse with it and get good answers back, without having to assume zero context in your responses. For example, you can ask it to write a blog post, provide some feedback on it and then it will rewrite the blog post based on the feedback (one paragraph in this post was written with it - reply if you can guess which!). Examples in the linked blog post also demonstrate it doing this with code.

With each release of a tool like this, we've seen a wave of products being built to support initial use cases: I wrote about this with Stable Diffusion, and the AI Avatar Generators that sprung up afterwards as an example.

What we're currently seeing is that these products are the ones capturing the majority of the value. Jasper, built on top of GPT-3, is an interesting example. Having recently raised $125m at a $1.5b valuation, based on $45 million in revenue in 2021 and the expectation of doubling it in 2022. Reportedly, the company makes more than OpenAI currently, who build the technology it leverages. While I personally feel Jasper is over-valued for the level of differentiation of their product (there are already other products that I've seen do more with a lower price-point), it is an interesting pattern that I expect we'll see play out in many other use cases as AI moves into more sectors and industries.

What is even more interesting is that it was recently reported that Jasper is now working with another company to build its own models. As the CEO claims, this is not to replace GPT-3, but to supplement where it is weak. That said, it does reflect another pattern that I've started to see: companies getting to market quickly using proprietary models, and then once they've reached scale, using open-source models and fine-tuning to offer the same experience (or better!) for improved unit economics, by building their own and becoming full stack.

If you've enjoyed this, please share it with others who you feel would be interested!