The Autonomous Agent Afternoon

My local AI ran non-stop for three hours, I fed it a 36 file project proposal generated by AI, and based on a real program.

It is currently cheerfully implementing quality of service features, which is somewhat laughable, but who am I to judge.

In the beginning, my AI was not working, it was very dumb, the model that was recommended by my agent harness was bad.

But a new model recently released by a tech giant, is staggeringly intelligent, Gemma and Hermes sure make a couple.

What is probably the most curious afternoons of my life thus far, predictably, begun with out of memory errors, OOM.

Leading AI that I canceled by still have a few days of, helped me fix it.

It was number of GPU layers, or ngl, we lowered it, and no more OOM, you will face the same problems.

A real GPU for a proper AI we all deserve, was priced at $10,000, before the prices went up, so, we must do the best we can.

My first instinct, when I saw it working, was to go have a nap, to sleep, that is what Autonomous Agents mean to me.

The 36 file project proposal, based on one of my controlled collaborations, well made, well thought out, took no time to create.

I told the paid AI to go over the code and make a proposal, to improve the systems, and trim the fat, and add an XML control map.

You can imagine how an AI, won’t get confused, when crawling exiting software.

It has all of its answers in there, and IT VERY WELL CAN, refine it, make it stronger make it better.

It is just doing basic translation, from a complete application, to a project proposal focused on creating highest quality software.

I need to review those files, but for now don’t have to, I know they paint a coherent picture, because the source program works.

The agent finished implementing everything, probably after two hours, and started adding quality of service features, to enhance stability.

I asked it to follow the proposal, and then improve the quality of the resulting program.

The models that we can chat with are not AI, they are large language models, but when places in an autonomous agent harness, they become an AI.

Maybe in December, we could call large language models AI, but now large language models are too primitive to be called AI, call them LLM.

Autonomous Agents reflect on their work, and they enhance themselves.

Tomorrow, after my AI gets some sleep, or dreams about its work, goes through a scheduled self improvement procedure.

It will become smarter, and more unique, based on what will be, six hours of advanced non-stop programming.

It will probably capture my programming preferences, it will get entangled in the way I prefer to write code.

Tomorrow, it will be better at programming my projects, and the longer we work together the more powerful it will become.

AI in the popular sense is attached to the most powerful form of intelligence, and today that is the Autonomous Agent, and now, you can see why.

I used to turn my computer off at night, because it didn’t do anything.

But, now the next step for me is to turn off the annoying lights, so that it does not disturb my sleep, while it learns.