Code I Wrote in the 1990s Is Still Running. Nobody Pulled the Plug.

November 02, 2026

Watch the video version:

Software I wrote in the mid-1990s is still running. It builds reports. It started on an earlier version of Visual Basic, moved to VB6 later, and kept an Access database behind it. I learned this recently from someone who still works at the company I wrote it for, early in my career.

Every technology it is built on is long past support and security updates. So they firewalled it onto a machine that cannot be patched. Data arrives on an incoming feed, the system runs its reports against it, and it has not been updated in any way in years. That plug should have been pulled years ago.

I was shocked. I did not expect it at all. I have nothing at stake in it. It was early in my career, I still had a lot to learn, and hearing that something I wrote that long ago still works was amazing. Then the harder question: how catastrophic will it be if it breaks, and there is no way to fix it today?

The common wisdom calls this a success story. If it isn't broken, don't fix it. Spend the budget on what is failing, and let the stable things keep running. Every technical leader has heard that line in a planning meeting, and plenty of us have said it.

The same pattern is already forming around AI. Gartner's 2027 CIO Agenda survey of 2,002 CIOs and technology executives found that only 31% have guidelines for when an AI pilot should not move to production, and only 17% have guidelines for abandoning low-value AI already in production. Launch criteria get argued over in every room. Exit criteria are the part that rarely gets written.

Here is what happens instead. The system ships, and it works. Running quietly reads as success, so it never gets a review date. Each year the cost of leaving grows while the system looks free. The people who built it move on, and what they knew leaves with them. I am one of them. That is what I call State Loss: shared understanding that resets because it lived in a person rather than the system.

Then the organization feels the risk and answers it with containment: firewall it, isolate it, don't touch it. That shrinks the attack surface. It also makes the system harder to change, and containment quietly becomes the plan.

The decision to retire it still gets made. When nobody writes it down, inertia makes it, and inertia picks the worst possible day: the day the system breaks, when the repair path is already gone. No supported tools to build it with, nobody left who knows how it works, and a machine that cannot be patched. A migration that could have been planned becomes an emergency rewrite, and it pulls a team off the roadmap for as long as it takes. In a system like this, the load does not even land on engineering first. It lands on whichever business unit loses its reports overnight.

An AI feature reaches that wall faster. If it calls a hosted model, that model is retired on the vendor's schedule, not yours, and every month it runs, it bills tokens. My reporting system took decades to reach the point where a failure could leave no way to fix it. An AI feature with no exit arrives there in a fraction of the time.

If you run a system like this, nobody on your team is in error. Every system that matters was either designed or it accumulated, and most accumulated. A system with no exit was never designed to end, so its lifespan accumulated too. The leaders who stay ahead of this write the exit on the same day they write the launch.

Exit criteria can be written before a system ships, from what you know that day: what it depends on and when support for each piece ends, what it costs to run, who can repair it, and what would replace it. Add a named owner and a date. The criteria are not a trigger. On that date, someone with standing evaluates the system against them before anyone pulls a plug. It can keep its place, and often should; keeping it just becomes a decision on the record instead of a default nobody made. Still running is not the same as still decided.

Pick the oldest system your team still runs. Answer three questions without looking anything up: when does support end for what it is built on, who could repair it this week, and when did someone last decide to keep it? Any answer that starts with "I'd have to check" marks a system that accumulated instead of being designed. Write its Exit Line from what you know today, and put the review date on a calendar before you close the laptop.

The Structural CTO™ is out today. A system with no exit is a system nobody planned past its launch, and this is the book about the seat that plans: reading where a company sits on The Load Curve™ and building for the load it has not reached yet, the discipline it calls Camber Architecture™. Get it today: https://TheStructuralCTOBook.com

The practices that carried a company to ten million are the very thing keeping it from thirty.

Anthony S. Jackson

Anthony S. Jackson

Anthony S. Jackson has spent 30 years inside technical organizations. He is the author of the Architecture Protocol Series: three books on the structural problems technical leaders were never told they would face. He writes the LeadershipOS™ Inner Circle, a monthly printed newsletter for CTOs and engineering managers who design teams that hold under pressure.

LinkedIn logo icon
Back to Blog