I* created my first web app in under 2 hours

*Actually Claude Opus 4.5, and it's 100% worth the hype

I created my first web app, called Bummer Currency. It is a web app for bottom-up allocation of tasks (e.g. household chores) among a group (e.g. a household) using a points-based system. Using the power of prices and the invisible hand, it ensures people do the tasks they enjoy the most/dislike the least relative to others in the group, making everybody much happier/less unhappy.

Here’s a screenshot that should mostly be self-explanatory:

To fully understand the system, it is inspired by the Putanumonit blogpost, which I cannot recommend enough, where the author describes a “bummer points” system they used during their military service to distribute tasks to great effect. Really, I recommend opening it now and reading it before continuing my post, to viscerally feel the power of the system.

It Just Works

Claude Code with Opus 4.5 is good. Very good. It just works. I created a fully functioning web app in just one to two hours. It is so good it is frankly misleading to say that I created the app: Opus 4.5 created it. I just told it my idea, shared the blogpost, and did the few steps (like setting up the backend Firebase) that it could not do itself. It wrote all the code in a few minutes or less. The part that took the longest was the steps I had to do, especially as I have not used Firebase before. But it was much faster with the help of Claude, as it explained every step of the setup process clearly and correctly.

There was one bug, and the AI told me exactly how to open the developer tools, what to look for and what steps to follow to debug. And, I’m going to repeat myself, it just worked. (The bugs are an expected part of Firebase, not a limitation of Opus 4.5. Something to do with indexing the databases when testing rather than deploying.)

If you are curious about the final result or how it was built, you can check out the links below:

• The Live App: Bummer Currency (Go ahead and play around with it. Warning that this is not intended as a final product but a hacky thing to work in my house. E.g. I can access and modify everything in the backend.)

• The Source Code: GitHub Repository

Reflections

First, it is worth the hype. At no moment in time did the AI go “off kilter.” It always did the right thing. It really felt like talking to a capable person who understood exactly what I wanted. My housemates already made a few feature requests, and each one took 5 to 10 minutes to get implemented with Claude Code.

Second, I believe we are about to see the software equivalent of YouTube. Just as video production became democratized 15 or 20 years ago, allowing anyone to make their own videos without a studio, it is now easy for individuals to make their own bespoke software.

Third, I also believe we are not far away from a single-person unicorn. Instagram had less than 20 people when it sold for billions to Facebook. People who can lead a team of AI agents can achieve amazing things in a short amount of time.

Fourth, I now believe the insiders at Anthropic more when they say they are on track to ‘a country of geniuses in a datacenter’ in 2027. It is easy to be skeptical and think this is just marketing hype. But it really isn’t. The team at Anthropic are going to be power-users of Claude Code, so everything they do is going to be much faster than anything else the world has seen before, and it is only going to get faster and faster.

Last, the scary part (if the above was not scary enough): I have not read a single line of the code. On the plus side, this illustrates how robust Claude Code is. I don’t have to read it, and honestly, I wouldn’t understand most of it as I have no background in front-end or back-end software engineering.

However, if the AI really wanted to, it could have easily done something dodgy. It could have inserted a security risk, or created code to snoop around my laptop. Or, maybe there are glaring security problems that Claude overlooked. Given the low-stakes nature of a chore-tracking app and the belief that Claude is more or less trying to do what I want it to do, I’m not concerned. Unfortunately, it is easy to imagine people blindly trusting LLM agents to write code for high-stakes scenarios without proper vetting, thus exposing themselves/the public to serious risks. Or worse, how easy it would be for a rogue AI to do what it wants given how much we will trust it.

What are you waiting for?

First, as I already recommended and if you haven’t already, read the Putanumonit blog post that motivated this app!

Second, you should experiment with Claude Code. The documentation is great and easy to follow. If you cannot do it now, create a time block in your calendar so you will do it. Really, I can’t emphasize this enough! Make sure you play around with this. It only costs £20 (in the UK) for a monthly subscription, and you can cancel if you don’t like it. £20 is a low price to pay, and who knows, maybe you will end up creating ten pieces of bespoke software that automate a whole bunch of annoying things in your life. If you have not heard of a ‘terminal’ before, then either have the chatbot Claude explain it to you, or instead play around with Claude Cowork, which is essentially the same thing but accessible to general public.

If you’re low on ideas, here are some ideas I am thinking of:

• automatically finding events happening in London given my interests and hobbies.

• automatically creating tweets, newsletter and slack posts for AI news I read

• doing AI safety research

And 13 creations from Dean Ball (I can recommend following their substack for insightful commentary and proposals on AI and AI policy):

1. Automated invoice creation, sending, and tracking;

2. Created scientifically realistic simulations of hydrological systems as a learning project;

3. Automated my research process of gathering and analyzing all proposed state legislation related to AI (though this is no substitute for reading the bill for anything I am going to write about);

4. Orchestrated a complex chain of autonomous data collection, processing, analysis, and presentation steps related to manufacturing and industrial policy;

5. Created a machine-learning model capable of predicting US corn yields with what appears to be very high accuracy (the proof will be in the pudding), based on climate, soil, Earth-observation satellite, and other data sources;

6. Replicated three machine-learning research papers and modified the approach to suit my own research ends;

7. Performed hundreds of experiments with Byte-level language models, an emerging interest of mine;

8. Created an autonomous prediction market agent;

9. Created an autonomous options trader based on a specific investment thesis I developed;

10. Built dozens of games and simulations to educate myself about various physical or industrial phenomena;

11. Created an agent that monitors a particular art market in which I am potentially interested in making an acquisition;

12. Created a new personal blog complete with a Squarespace-style content management system behind the scenes;

13. Other things I cannot talk about publicly just yet.

Or check out Privacat’s detailed experience of creating their own RSS feed and pipeline to triage the posts in it. (Can also recommend following their substack for a lawyer’s perspective on AI.)

So, what are you waiting for?