Programming With AI For Young Ladies, Your First Day

Yes, I finally beautifully failed enough times to satisfy my curiosity, and I sat down to creating what AI actually needs to program your software.

I must be brief, because this is too important to muse about.

Frost of all I needed program that would serve html files, and I simply asked in chat for an http-server, pure node, that opens firefox.

No configuration required, just a simple request.

But, to create something that has multiple parts, you can’t just use chat, multiple moving parts cannot be created in chat.

You start by tidying up your main AGENTS.md file, and there I said I prefer ECMAScript Modules, no npm dependencies.

Then I created my project folder, and had AI write me another AGENTS file, these are combined together, this one is project specific.

I added two weird things, they are called SKILLS, which define steps that need to be taken.

I asked AI to keep a New Line Delimited JSON blog, that it will show in a minimalist little bird tweet style, always with a diagram.

And in my AGENTS I said, use this blog skill every time we finish something, the result of this is codex locally blogging in human friendly terms.

And I added an iffy idea that I called CRFC, RFCs are not AI relates, and they do not define steps to be taken.

They define what a thing is supposed to be, the C in RFC, is short for compact, that is something I added.

As as a result I am able to say, make sure this thing is RFC-0009 compliant, if it is not compliant, research, use your SKILLS, use your AI brain.

A lot of people will like the blog idea, especailly those who will understand it, as a user interface for artificial intelligence work.

But only ancient warriors will like my CRFC, thought no doubt they had a batter idea, maybe even, no standards, skills only.

I should also mention that I keep my npm and other software in a docker, it is a lightweight virtual computer that protect my computer from attacks.

I then gave my AI a set of images and said make a theme, and it created a very primitive flat color user interface theme with buttons.

And that is actually correct, it used AI vision and it did everything right, and did not try anything to make it perfect.

It made 4 blog posts, and this is where I stopped the project, this is day zero, and it is perfect.

Now, you do not speak to this AI with your own words, think of it as a live thinking machine that will do anything you need done.

You need to stop speaking to it, you go back to your smartest AI in chat, and carefully and wisely, with an open mind ask:

“What would it take to add this super complex thing I need it to do?”, and when you get a doctorate/phd level answerer you ask for a SKILL file.

And you give your coding agent that scientific skill, and then you tell it to use that skill wen creating your thing.

You see, your words and requests are weak, you need to forge a skill first, and then mention that skill in your request.

Not only are you building up a skill library, not only is it stable and repeatable, not only can it help you meet a CRFC standard if you are into that.

But it is an exceptionally clear request, that will do the thing you need done.

If you ever find yourself in a situation when something is bad, it means you did very badly, in communicating with AI.

Notice, how the chat does one thing and one thing well, the manufacturing of a single coherent SKILL.

And now you can ask your coding agent to use a string of skills, and create a new SKILL, based on existing SKILLs.

Or to keep it clean, compress your skill library into a zip, and send it to your chat AI to make new composite skills you need.

Talk to your chat with your own words, but your coding agent is a different kind of thinking machine.

First enable it with precision, and then order it with great seriousness and gravity.

Chat is your buddy, the programmer is a serious scientist.