The Problem With Never Deleting Anything
Think of your home garage. You keep adding tools, boxes, and old furniture — but you never throw anything away. Eventually, you can't find the things you actually need, the space costs money to maintain, and some of that old stuff is a safety hazard. Software infrastructure works exactly the same way.
In many organizations, Platform Engineering focuses entirely on building new things. But the most dangerous technical debt is often the old, unmaintained tools that refuse to die. We call this Zombie Infrastructure: systems that are technically still running, but whose original owners have long since left the company, whose documentation is missing, and whose purpose is unclear.
Zombie Infrastructure is expensive in three distinct ways:
- Money: It consumes compute resources — cloud VMs, storage, and network egress — 24 hours a day, whether anyone uses it or not.
- Cognitive load: Every on-call engineer has to mentally account for it. Every new hire has to learn what it is and whether they should touch it.
- Security surface: Unmaintained systems accumulate unpatched CVEs (Common Vulnerabilities and Exposures). A zombie service running an old version of OpenSSL is an open door for attackers.
Deprecation is Engineering
Deprecation is not just sending an email saying "Please stop using X." It is a deliberate engineering process that requires the same level of planning, architectural design, and empathy as building a new feature.
A complete deprecation strategy has several components: discovery (who is actually using this?), communication (how do we tell them, and how often?), migration (how do we make it easy to move?), enforcement (how do we make the old path painful?), and termination (how do we safely pull the plug?). Each of these is an engineering problem, not an administrative one.
If you don't engineer your exit strategy, you will end up maintaining 5 different versions of your deployment pipeline forever. The Platform team becomes a museum curator instead of a force multiplier.