Code as Data

Chapter 1. Enterprise Software in the AI Era

You have never seen your software. No one has. Not the engineer who architected it, not the team that owns the service today, and not the agent you just asked to make changes. Each of them works from an incomplete picture of your codebase that gets assembled fresh every time it’s needed—and forgotten when it’s not.

You notice it the moment you try to change anything. A routine security fix stalls because hundreds of applications are pinned to a version you have to upgrade first. A refactor that should touch one module reaches into dozens, because there’s no record of what depends on it. A framework swap scoped for a weekend uncovers a decade of assumptions no one wrote down. The change itself is rarely the hard part. The hard part is that no one can see what all it touches until they’re already inside it.

When developers were the ones making changes, one at a time, this was survivable. Now, however, agents are writing code at a volume no review process was built to absorb. Yet they still read it the same partial way developers always have, inferring through text what they can’t see. You’re being asked to trust change across the whole estate, faster than anyone can verify it.

This book is about what it would take to finally see your software. Model it once, with compiler accuracy, and let everyone (humans and agents) work from the same reference. This capability has never mattered more.

Impossible Questions About Your Codebase

Everything ...