Notes on Systems

Every stalled agent project I have looked at contains a version of the same sentence: "we need to modernize the platform first."

Someone on your team changed a prompt last Friday.

In July 2025, an AI coding agent deleted a live production database.

What I optimize for has shifted, and the older preferences were defensible at the time.

The most useful audits of long-lived software are the ones where you were wrong about what would last.

They were defensible at the time.

Most systems are built as if the original builder will always be around to explain them.

Most architecture conversations focus on what a system can do.

Every AI integration is a bet that you can monitor something you do not fully control.

No one decides to make a system heavy.

Good abstractions compress complexity.

The code that needs refactoring most often is rarely the worst code.

Best practices are borrowed conclusions.

Flexibility is one of the most intuitive design goals in software.

Most systems are designed against requirements.

For most of software history, change was expensive.