“If you can't explain it simply, you don't understand it well enough.” --- Einstein

We all know by now that AI can write programs, and it works best on small well defined parts of a program, or little bits.

So what I bring you here, is a framework around just that, that helps you create programs a little bit at a time.

The novice does not need as much help as it would seem, for one the AI can explain how a program can be structured.

So as long as we tell it what architecture to use, and the simplest one is a pipeline or a tree for envelopes of data.

Old banks and office building used to have that, you jam an envelope into a shuttle, and send it to a specific department.

That department will do something with it, and either pass it on, of sends something else in its place.

So here a department or a person in that department, transforms the data in the envelope or shuttle.

We replace that person, with a program that is written by AI.

The network of pipes, is just a visual programming language, boxes that represent programs, box ports that represent input and outpus.

And pipes, that connect output ports from one box, to input port or another.

This would work, already, but it would not be impressive, programmers that do coding by hand, would out-code you.

And we can’t have that, because just, who do they think they are.

We are going to upgrade our box, to the concept of a supervisor, that uses a worker, in supremely clever ways.

If you are downloading a file, your supervisor will manage failed downloads, but if you are downloading many files…

The supervisor will manage multiple simultaneous downloads, and another supervisor that ensures that file is downloaded.

Nothing too special, but there are many kinds of supervisors, one of them is a self provisioning remote program runner.

Meaning you can automatically scale to handle extreme workloads, but temporarily renting more computers to execute our little bits.

Some supervisors will be meant for executing code on CPU cores, others as standard web browser workers, as UNIX processes, etc.

But I wouldn't call this self healing program architecture yet, and you might have guessed, yes, Artificial Intelligence Supervisors.

Adaptive, intelligent, program execution, and very high self healing potential.

Now, out standard, average coder, is a little worried, as there is a lot of moving parts here.

It would be hard to make all this go, without a Visual Programming Language.

So we got something here, you are becoming powerful, beautiful and a little bit scary.

The beauty comes from design standards of the Visual Programming Language, one of the visual themes will be inspired by the now retired Apple Dashboard.

Except Generative Art AI will make the widgets even more beautiful, The visual programming language will have a Ninja mode as well, so it is OK.

So AI tells you how to structure a program, write your little bits or functions, supervises their execution within safe constraints, and generates a beautiful GUI.

It is important to know that the concept of a supervisor, comes from the Actor Model of programming, a well researched beast.

And one of its strengths is the Queue, meaning all the shuttles or envelopes sent to the supervisor, go on the queue.

That is where the worker(s) get their jobs from, they scan the queue, sometimes local, sometimes a remote database for jobs.

This in turn translates to a wonderful development experience, as you don’t need to have the full program, to do the first part!

For example, you are tired of crappy User Interfaces, and decided to scrape your favorite websites for news and smart comments.

You will run everything through AI, to to expand, maybe research, and pinpoint important comments, and perhaps summarize.

But all you need to get started, is to create a Smart Supervisor, and tell it to scrape a web page.

Here you add a test to see if the article content is extracted in full, the AI will keep the queue of raw data to be parsed.

And will probably do the scrape on the first try, but you will still run the test, and AI will retry, and refine, if it fails.

With one little shiny box, the AI extracted and formatted the data for you, and will keep it in the queue while you are building your program.

Until you save that data, to where it needs to go.

Until then, you can relax, reason about the data, think about your next move now that you completed your first, that is rare.

Even more importantly than that, the AI, will have no trouble, noticing a parsing failure when your website changes…

And it will automatically know, to rebuild parts of the scraper.

Now, our manual programmer, still unpressed, feels a heat sensation in the back of their head, it is a hard worry bubbling up.

And here, we will finish them off, with one final trick, now a psychological move, that they are incapable of.

Because, now you get up from your desk, without documenting anything, and without even a singe thought about the program for another two weeks.

You just get up, and leave to do more interesting things.

And when you return, you will instantly visually understand the program, and the AI will be there to assist you, with all your needs.

The supervisor architecture ensuring that you can efficiently scrape 5 websites, or 10,000,000 if you can afford the self provisioning network bills.

Now our programmer, witnesses how very human the design is, and defeated, it leaves, and I for one hope they get some rest.