Death of Software

Software development is changing.

We know this, we've discussed this, it's been beaten to death.

Why write about it again? Answer: Because why not?

Sit in silence for 10-20 minutes. Stare at the blinking prompt. Do not type. Just feel. Maybe think. Let your human context window become saturated with the problem. 

Then, and only then, once it’s physically impossible not to hold back the flow of thought, begin dictating to the computer. 

Don’t structure the thoughts. That comes later. You’re brain dumping. Speaking things that you previously only dared whisper before, but now… anything is possible. 

Then let the genie do its work. It will sort through your brain dump, your whispered hopes and dreams, your imagination. 

And then, it will extract some ordered meaning. But it won’t click for you. It won’t feel right. 

You ask it to draw you a picture. 

visually explain this - I don’t understand the data flow that we’re trying for here

Let it cook again. 

Finally, it finishes. You get a pretty picture. A mermaid diagram , with data flow, call trees. Something that looks like a mediocre lucid chart diagram that your L4 engineer self would have thought was amazing work. 

It’s been 35 minutes since you first sat down and stared at the prompt. 

The tool shaped object has molded to your hand. Perhaps, some might say it is a reflection of yourself - your hopes, dreams, your soul. It’s a dirty mirror or perhaps a funky fun house mirror, but it shows enough. 

codex —-yolo “/review the plan in git status”

Codex is going to take a long time. that gpt xhigh thinking is slow. 500 line markdown file can take 20 minutes to validate. 

And yet - you wait. 

You wait not because you want to, but because you’re starting this process over in another session, another project. 

DING

Codex is done. You switch back to it, and discuss the feedback on the plan. Some referential integrity issues. Inconsistency in data interfaces. Overly complex rules engine when a guard statement or type check will do. 

You go back and forth a few more times. Maybe you’ve built a tool that helps you with this, and it knows your style and design principles. Maybe not. 

You’re ready to turn it into a PRD. But what even is that term anymore? Product managers did PRDs. They interviewed dozens of people. Consulted experts. 

You ask it for diagrams. For mockups. You tell it that color scheme doesn’t work, it’s too bright. You want the UI to inspire calm and, be soothing. 

You go through 5 or more rebuilds before it clicks. The tool shaped object has created… the tool. 

The job to be done can now be done. 

And you feel the rush. 

But it also… feels empty. Because the part of your soul that takes pride in how you spent so many hours cultivating your skills… cries out. 

It cries out that you’re doing it wrong. That this is cheating. That whatever happened wasn’t real. It wasn’t authentic. It was… artificial. 

And you ignore it. Because you’ve got the rush of the future. Of embracing what is to come. 

But, in your heart, you still yearn for the past. You yearn for modernity to return to the simpler times before. 

DING

Your other agent is done. It’s time to begin again.