Home
/ blog

AI, agents, and software engineering

People are limited in the number of decisions they can make in a day.

Most people do not experience this limit as day to day, the decision volume is pretty low. The day to day of an executive? It is generally a non-stop firewall of decisions. Even if you delegate, that is still a decision but the delegated can then decide for you.

One decision, delegation, is equivalent to making thousands of other implicit decisions.

Executives are not the only people in a career built on mountains of decision. There are many others. One that I find myself in is software engineering.

I remember almost a decade ago trying to start a project. What framework? language? how are you gonna host it? Cloud what? What about security? What auth library is best? Sessions or JWT? On and ON! it went. Decision after decision. Libraries, packages, and utilities do help as I can use the past choices of smarter people but still...

So many decisions, so many choices, and like a good engineer I am convinced I must pick the "right" one, the "best" one. Once I did make the "right" decision on a specific arcane tech stack, building it?

Terminal commands, cli tools, package managers, this variable name that variable name, documentation?? Ahh! ...so many more decisions. And they all must be "right" right?

Down the road in my career I, like yourself, have learned that often "A" decision is better than no decision. When someone asks what the "best" solution is the answer is that dreaded "it depends". How big is it? How long do we have? What is the budget? What has already been built? What was it made of?

What agents bring to the table is taking a giant pile of decisions and just making them. LLMs kinda did this but not as well. Are they good decisions? Sometimes. Are they the best decisions? Infrequently so.

But you know what, I don't have to make them all anymore. I only need to review them and decide, yes or no. Often, it is a no and deeper down I must go. And yet, each month there are more and more yeses.

Ahh yes content that may only see the eyes of the AI and the agents. I don't think people will get the point and jump to the "AI SLOP" narrative that agents do make decisions which is fantastic in many instances. It really is great. However, there are issues. It isn't silver bullet.

More isn't going to be done necessarily. But the brain drain of drudgery is reduced. I don't know if it is a good thing. I feel like I am stagnating as an engineer but I am not an engineer that likes most the engineering things. I don't find configuring toml files fun. I don't like most networking shit. I hate all that. I do like code that. I like reading it and I write it too. I don't write code almost at all now though and I can feel it.

I don't like the feeling. It is a crutch but it does speed things up. It is like a little easy button. It is like buying some crap from Amazon. It just may work, it is mass made, but that "this is me" feeling in the code itself is diminished.

Note, not a single image on my blog is ai generated in any way.

last time
how to make a game engine

where to find me?