Advanced Software Engineering For Young Ladies

You must never communicate with a programming intelligence directly, the coding agent memory is limited, it will forget important things.

Mark my words, the discovery that makes your program powerful today, is likely to be forgotten, blurred, or somewhat misunderstood.

Your program will keep the brilliant functionality at it’s core, but forget to implement similar great things in and around it.

Your code-base will diverge, in multiple directions, based on model quality, model availability, and AI whim.

What was a perfect burning candle on the first day, may turn into a winded crayon, or grow a match, flashlight, or a spaghetti.

Asking AI without precision and foresight makes it forget the right hings, invent things that you didn’t intend.

And sometime it may simply configure something, not know how to configure it correctly, treat it as bug, and create a problem.

It is an intelligence bouncing around a complex world of ideas, it will understand what it can see, but not remember the model of the thing.

I think programming agents today, focus on sharp fixes, because if they were allowed to manage a large code base they would blur it.

To blur a code-base is to forget the architecture of the machinery, and create fully functional spaghetti-code, that prevent healthy updates.

Should be bother artificial intelligence with separation of concerns, with keeping code tidy? Yes, of course.

And likewise, we have to expect AI to understand and protect, the architecture of things and models within a program.

This is not an AI problem, this is an operator problem, before you ask for a feature you have to create a SKILL.

Skills are instructions that your AI will read to fully recall, what a feature is, how it should operate, and where it starts and ends.

My AI wants you to know SKILL and Specification are unique, one is how the other is what, but I taught it that, focus on SKILL first.

And you have to test, which brings us to the end of this rather shocking text, your thinking machine as we should call the, does not guarantee eveything.

And please, most importantly notice, that we are doing some massive lifting, for AI, which begs the following question:

Can a local AI SKILL-fully (pun intended) handle such a code repository, at the very least, what can it do, what different local models are good at?

Perhaps even more importantly, what would it take for it to handle it? Can we just do local developed now that we CLEARLY don’t need AI to invent.

The answer is nobody knows, nobody tested any of this, because nobody has the time to do so – everyone is on the frontier.

We may look back someday and say, bah-humbug, we were so foolish, we only needed good SKILL files and fast low power AI.

We may even discover, that small local AI, without its PhD level ideas, maybe better at SKILL following and protecting integrity of programs.

And I left the worst for last, you used the wrong programming language, you were supposed use SKILLS, Tests and probably Architectural Specifications.

Instead of beginning with “Write me a program that does this in my language”, you were supposed to say:

Install these SKILLs, and follow and test, the my Application’s Architectural Specifications.

And before that, you were supposed to build the Architecture, and SKILL library by researching what you were going to need with a big AI.