Launched my first web app that I built entirely by myself with AI
Magicdoor.ai has been live for a month, and I now know how to build stuff with NextJS
About 3 months ago I started learning to code with AI. It’s been one of the most fun and rewarding things I’ve done in years.
Here is Magicdoor.ai, and you should definitely check it out if you haven’t yet. Magicdoor is a wrapper around the most popular LLMs, and allows you to use them all from one place. It solves a personal pain point for me where I was paying $60 per month for AI at one point. I use ChatGPT, Claude, Perplexity and I also like to generate images and play around with the latest offerings for all of them. I’m a heavy user of AI, but for my day to day usage, I just want something that works.
Magicdoor is $6 per month + whatever your actual usage is which will be much lower than you probably think. So far, my usage has been about 50 cents per week. Here are some examples:
A single, short question for Perplexity (Best):
363 total tokens
$0.0005 per message
$0.0068 total cost
A longer chat with Claude 3.5 Sonnet to learn about SEO best practices:
3,020 total tokens
$0.05 total cost
So, make sure to check out my work and pay me money!
Hopes and dreams
For the first time in my life, I am an early adopter of what I believe to be one of the most transformative technology/platform shifts of my life. I was too young for PCs, disinterested in building for the web, and I missed mobile because I was a banker back then. But the timing of AI can’t be better. I’m young enough to really get it, and old and experienced enough to do something about it.
But even from my vantage point, reading voraciously, using AI daily for both work and private stuff, I find it impossible to predict the future. I have no confidence in my ability to make the right, single, bet, raise money for it and go all-in on one idea. The landscape keeps changing and OpenAI and others regularly pull the rug out from under complete categories of AI startup.
In the web/mobile startup era, people would always ask “but why can’t Google just do that?” It was an annoying question, because they usually could, but they rarely did. And also, yes every startup comes with wipe-out risk. Most startups fail, and usually not because Google killed their market. Google couldn’t go after every market. But in the AI era I believe this is different. OpenAI has a vibrant consumer business, and they do develop it rapidly. They actually do constantly build products on top of the core tech. It is therefore that I believe there will be few unicorns in the AI wrapper space. But I do believe there is a huge opportunity to build small, profitable, ultra-lean products. But what are they?
There really is only one way to find out. So my conclusion was that rather than trying to come up with a perfect idea that I have great conviction in, I just wanted to be shipping and launching stuff into this crazy environment.
I have launched! And I am shipping:
I’m so happy about this
The first time I’ve wanted to build and launch small online businesses was in 2013, while I was working at AlpInvest. But I had no idea how or where to start. For over ten years I have painstakingly collected all the experience points I needed. At Rocket I built teams, did sales and marketing, go-to-market, and some product. When I was living on Bali I built Carbonfoodprint.life and a Wordpress website for my consulting biz. Then I built Ox Street, where I did much more go-to-market, company building, and picked up UI/UX and some basic technical knowledge about how software is built.
But the experience of launching and building Ox Street was frustrating and difficult. I hired some engineers who turned out to be in way over their head. Everything I wanted to build I had to explain to a designer, who then took many days to design it. Then it had to be explained to an engineer, who took many days to code it. Then there would be bugs. Every move is just so damn costly. That’s why I taught myself design back then, I had to take one link out of this chain. But the dependence on engineers was still a thorn in my side.
Now, finally, I have gained the ultimate ability for the ‘idea guy’ that I’ve always been: I am now able to launch an minimum viable product all by myself.
It didn’t have to be this way. I could have continued having lots of cool ideas, continued being opinionated and often right about where the world is headed, and continued not having anything out there that actually does something. But that’s not what happened, I just fucking started building again, and now I have made $100 in Gross Profit from my new app. A lingering sense of guilt and shame I’ve often carried with me for not doing ‘enough’ with my ideas has disappeared. I’m very proud of myself, and that’s a great feeling.
How nice it is to bootstrap
I haven’t invested any money to launch Magicdoor, and it’s been profitable in the first month. My total costs are about $40 per month now, and at 11 paying subscribers, my revenue is $66.
In contrast, with Ox Street I had raised around 325k initially, and by the time the product launched there was less than 150k left. That kind of time pressure drives an intense, frenzied work ethic that has benefits. But it also forced my hand into a number of decisions that weren’t necessarily good. What a difference it makes, not to be sitting on a ticking time-bomb but instead to have all the time in the world.
Plans for Magicdoor
I’m working on adding image generation models, and then I’ll start working with the ‘Assistants’ API. I’m targeting regular AI users who stop short of being full-on nerds. They’ll know what Claude is and find it noticeably different from GPT-4o, but they don’t have time to care whether they use claude-3-5-sonnet-20241022 or claude-3-5-sonnet-20240620. I’m targeting people who use AI to get things done, not necessarily those who want to endlessly tinker with it as a hobby, or those looking for a virtual girlfriend. My plan is to launch specific use cases, and try to launch lots of them, all rooted in problems I have myself and things I discover AI can do very well:
Shazam for everything: identify things from a photo. Recently on vacation me and my family used AI to identify a butterfly, a snake, and an open-pit mine I saw from the plane
Content writer trained on examples: For Carousell I made a custom GPT that writes in the way we want communicate to users. It can translate policy documents into user facing helpdesk content for example, and it’s amazing
Calendar tools: Create a calendar event from any type of message or screenshot (like a flight confirmation or a text message confirming a hospital appointment)
Sentiment analysis and tracking for brands
Competitor intelligence
Management coach
Etc, etc, I have a list of more than 20 mini-product ideas!
One potential future is that Magicdoor exists as a sort-of test-bed for mini-products and that I’ll spin out the most successful ones into their own brand and dedicated app. But Magicdoor itself could also evolve into a more and more interesting tool. I think of it as a kind of an anti-fragile platform play. By being multi-model, multi-modal, and multi-channel it should be somewhat protected against rugpulls.
But first, I have to create a stop button. I made a typo yesterday and had to wait for the whole AI response to finish before I could retype my question. Other chat interfaces have a stop button for this reason and Magicdoor also needs one.
Let’s see!
Congrats on the launch and achieving profitability, Gijs! Your point about bootstrapping and cost is why I love software. It's so easy and affordable to build software nowadays. More people should be building :)