AI
AI coding agent is like the cat.
Because the conversation resembles something that is alive, with some of the personality of the cat. A hint of something intelligent going on. And not entirely predictable.
That is, of course, an illusion. I know that, you know that. AI is not alive, like the cat.
What the cat and AI has in common, though, is stupidity.
For being a cat, it is quite smart. It has learned which of the humans are safe and nice, who brings the food. It knows I am the one it can bite. That doesn't mean it is smart, just that it has the capability to figure out the essentials for survival and a nice living.
The complete lack of ability to reason and logic. No line of thought, no regard for consequences. That is the nature of the AI. The cat is different, in that it has some resistance to doing things that may cause harm or simply does not seem beneficial. It has some line of thought, it predicts the outcome of actions and weighs this outcome against potential benefit.
Now, I can never tech the cat to code. Not because it is too stupid, it is, but because it sees absolutely no benefit in following any of my instructions. There is no motivation.
Focus on the outcome.
Just describe the outcome, when prompting AI.
There is no need to to try to outsmart, to try to find your own, unique method that no one else has.
I found out a while ago, when working with coding agents. The outcome is what matters, and that outcome is what is most important to describe in the most precise and exact way. The process involved to find the answer is not up to me who ask, it is already determined in the way the agent works.
This guide from ChatGPT explains how to use their service efficiently, and to no surprise, it does not require obscure tricks to try to circumvent shortcomings:
https://learn.chatgpt.com/docs/prompting
Expect AI services to evolve in this direction, to improve to get you the outcome you ask for, with no other information than what you want and only add sources relevant to the question at hand.
Service providers continue to improve, they want their services to be as good as possible, and that means easier to use to get a desired result.
Coding with AI was the last to be attacked. We now see repositories banning AI generated code.
But bad code is the result of bad AI usage. If you use it wrong, you get wrong results.
AI automation is great for coding, when used properly. It automates and speeds up coding tasks, saving time.
Here, as with AI generated slop content, it is the desire and wishful thinking of creating without effort. The idea that knowledge and skills are no longer necessary.
On youtube, many genres are now effectively destroyed by AI slop. Those videos are annoying because they act like noise that drowns real content with real value.
Now we see more and more software apps generated by AI. Usually of poor quality, because the AI generated code does not work well, and the developer has no clue how to fix things.
We no longer want AI.
A result of wrong use of technology.
But software automation with neural networks and LLM models works, within its limitations, when applied with sense.
For a software developer, even creating applications that does not actually work, has value. Because it can still be used to see what a specific type of application usually has inside, I can look at the code, see which interfaces and protocols I may need.
I can upgrade a software to support newer versions of a framework that application depends on.
I can migrate old code to a new architecture, fully automated. Yes, it works afterwards.
Translation. Including transcription from a video. It works.
Image processing, with detection and classification of objects.
Training control algorithm parameters, such as finding feedback matrix coefficients for dynamic systems.
Manageable, smaller tasks that improves productivity, saves us time.
But none of that require large data-centers. Those AI systems run locally, on my computer, on small, dedicated models tailored to solve just one specific problem.