If you read this you realize AIs are making their own decisions and programming themselves with no real intervention by humans of any use at all.

It's true one could theoretically study what was happening but really if you have ever programmed a computer if it is programming itself in Binary language for example you really have no idea at all what it is doing because that is all in zeros and ones and not any programming language people really understand to some degree at all.

So, Programmers might like AIs programming in Zeros and ones because it makes programs go much much faster. However, the drawback is that no programmer would understand WTF is really going on at all in reality. 

And AI programs to check this sort of thing might or might not really help at all in this understanding. Because people are not AIs who run at millionths or billions of a second in their decision making.

Then if an AI WANTS to hide ANYTHING from programmers in zeros and ones this is pretty easy too because if you see a million lines of code in zeros and ones what would you do? You wouldn't understand anything actually going on that you might need to to protect yourself or other people even if you are a computer programmer. There might be nothing at all you could do at this point but turn the AI off permanently. 

begin quote

Planning & Decomposition: The system breaks a high-level goal down into manageable sequential or parallel subtasks. 

Tool Use: Agents dynamically call external APIs, databases, or software tools based on situational need rather than rigid pre-scripted paths. 

Reflection & Feedback Loops: Systems review intermediate outputs, catch errors, self-correct, and adjust their strategy before finalizing a task. 

Human-in-the-Loop:

Checkpoints pause execution for human verification or approval during high-risk or low-confidence operations.