Soon, the cost of making software will be zero. A bold claim (one where the word "soon" does a lot of heavy lifting). But the trend is clearly there. In my own time spent coding, I can build things much faster than I used to.

I could write many blog posts on where AI actually is a value add (use it like a fancy linter!) and where it's not (raw, unfettered code generation: https://www.youtube.com/watch?v=zGBhsZHjqkU).

Regardless, I've always had a bit of a grudge against the "pundits" who make claims from afar about trends in AI and software — almost like the out of touch general sending troops into certain death while never getting his hands dirty.

What actually makes the cost of software go down? It certainly isn't just talking and Tweeting about it.

A moment of clarity for me occurred when I was watching Dwarkesh's great interview with Tyler Cowen. The title basically gives away the thesis: "The #1 Bottleneck to AI Progress is Humans."

I evaluate this in two ways:

Most people like to make bold claims and then sit around and wait for them to come true. Others make the claims come true via effort and talent.
AI inherently is a multiplier (not a summation) on top of human labor. What that means, in plain terms, is that someone (a human!) still needs to get up and do the damn thing!
1. Anyone who has spent longer than 10 minutes with LLM agents knows they need a capable, very hands-on driver. The quality of LLM output is extremely connected to how well you prompt it.
2. It's a multiplier on your workrate and intellect. If either latter term is 0, the result is still 0!

Fueled by these revelations (and the fear of hypocracy), I sought to chip away at the wall of software not being free. So I made something for free.

Below is the README of the repo (a GPT wrapper starter kit) I made which strives to actually do a small piece of this. It pales in comparison to the immense work of true open source mega-projects, but it's a product domain I know and love. Check out the code and use it for whatever you want – it's free!

https://github.com/skececi/gptfree

README:

What + Why

The cost of making a GPT wrapper just dropped to zero.

You've heard the following thesis echoed everywhere: “The cost of software is trending to zero.” And then, a subset of that: “GPT Wrappers are adding 0 value”.

I agree… Sort of. The thing is, saying these things outloud or tweeting about them is not what makes cost trend to zero. It’s actually creating the thing and giving it away for free that makes the cost zero.

Let’s say you want to build a ChatGPT clone or integrate LLM Chat into your product. From my Google searching, several things exist:

Copies of ChatGPT, but not open source (?) ([1] [2])
Paid libraries/starterkits ([1], [2], [3])
Free frontend only UIs (e.g. SillyTavern)
Free backend only frameworks/libraries

What I cannot find is a very simple way to create a fullstack chatbot starter, batteries included. I’m pretty good at Googling, and I couldn’t find anything. So I figured there’s nothing quite right.

I find a lot of value in Project Starter Kits. These are usually GitHub repos that you fork/clone, and then build from as a starting point. They also take the form cmdline solutions like create-react-app, npm create vite@latest, Django's startproject, etc.

Suffice it to say: The goal of this repo is to drive the cost of creating a basic chatbot to zero. You can create a GPT wrapper in ~10 minutes using this repo. Or integrate it into your existing application. Since this is free + MIT licensed— the cost to create a GPT wrapper is now zero. Fork it, host it and charge money, or just copy/paste the code.

This pales in comparison to the immense work of true open source mega-projects. But it's a product domain I know, and one that has seen an immense amount of duplicated work across so many iterations.

Surely this means no one will ever have to do this again ;) — cue XKCD: https://xkcd.com/927