People forget, but organizations forget at scale. AI is helping to plug that gap.
Welcome to Lead Prompt // executing leadership from the root. I’m your host, John Collins.
People forget. It is an immutable part of human biology. But organizations? Organizations forget at scale. And when you are operating in a complex technical environment, that collective amnesia isn’t just a minor operational annoyance, it is a critical architectural bottleneck. Lately, artificial intelligence has been stepping in to plug that exact gap, and a recent migration project with my engineering team brought this exact problem into absolute focus for me.
Recently, my team and I were tasked with moving a legacy application from Windows 2003 to Windows 2019. Now, on the surface, this might sound like standard life-cycle maintenance. But the migration was an absolute operational requirement to move this old application to a brand-new domain controller. Why? Because modern domain controllers flatly refuse to support Windows 2003 hosts trying to join the domain due to updated security protocols.
The application in question is at least 25 years old. To put that in perspective, the most recent comments left in the code are from engineers writing updates in 2003, all of whom have long-since left the organization. My current engineering team had no idea how it worked under the hood, or how we could safely lift a quarter-century-old application from a 32-bit Windows 2003 architecture and drop it into a modern, 64-bit Windows 2019 environment. Oh, and the cherry on top? The business ideally didn’t want us touching the core source code or executing a recompile. Fun times.
This harrowing journey down technical memory lane got me thinking about how organizations "forget at scale." It is a form of organizational dementia. The older a project becomes, the steeper the curve of institutional forgetfulness. Last year’s legacy application is a walk in the park compared to last decade’s.
So how do we handle such profound system amnesia?
Before we look at the modern solution, we have to look at the science of why this happens. In the field of management science, this phenomenon has been studied extensively under the label of organizational forgetting. We spend an enormous amount of time talking about organizational learning: how companies acquire knowledge, optimize delivery pipelines, and scale intellectual property. But academic research reveals that forgetting is an equally powerful, sometimes destructive force.
In a landmark study published in Organization Science by researchers Pablo Martin de Holan and Nelson Phillips titled “Remembrance of Things Past? The Dynamics of Organizational Forgetting”, the authors establish a foundational framework for how institutional memory breaks down. They separate organizational forgetting into two primary dimensions: intentional and unintentional.
- Intentional forgetting, or "unlearning," can actually be a healthy strategic asset like intentionally purging obsolete routines to make room for new operating models.
- Unintentional forgetting, however, is a silent organizational killer. It manifests as "knowledge loss" when key engineers walk out the door, or "asset depreciation" when the organization simply stops interacting with its own historical data repositories.
Furthermore, landmark quantitative research on manufacturing and corporate environments by scholars like Linda Argote has proved that corporate experience doesn't stay fresh indefinitely. In fact, knowledge acquired during production and development depreciates surprisingly fast if it isn’t continuously reinforced, documented, or embedded directly into everyday organizational routines. Without constant rehearsal, the organizational cues that trigger corporate memory disappear entirely. When you combine standard employee turnover with a total lack of continuous code interaction over two decades, you are left with an absolute cognitive vacuum.
In software engineering, this amnesia is uniquely toxic. Code is a perfect, frozen snapshot of historical logic, but the context around it, the why behind a specific hack, the underlying infrastructure dependencies, the environmental quirks, all evaporates completely.
When we started tackling this Windows 2003 migration, we fortunately had access to the original source code. It consisted of a classic ASP web front end calling Visual Basic 6 COM+ objects. Standing before that wall of ancient syntax, I was reminded once again of a saying a brilliant previous colleague of mine used to utter whenever we hit an undocumented feature or an opaque error: “Use the Force, read the Source.” He was clearly a massive Star Wars fan, and his philosophy was simple: the ultimate truth is always found in the code, not the outdated corporate Wiki.
But let's be completely honest with ourselves. In the past, "reading the source" of a 25-year-old application meant weeks of tedious, manual code auditing. It meant dragging senior developers away from high-leverage roadmaps to act as software archaeologists.
Thankfully, unlike in the early 2000s, today in 2026 we don’t have to read every single line of historical source code alone. We have advanced LLMs to do that for us, working exponentially faster than any human engineering team and operating at a massive scale.
For our migration project, we decided to train Claude and Gemini directly on the legacy codebase. We fed them the entire directory of VB6 classes, the classic ASP scripts, configuration files, and network schemas. Then, we initiated a rapid, iterative troubleshooting loop. Every time our QA environment threw a generic 32-bit memory allocation error, an unhandled exception, or a permissions failure on Windows 2019, we didn't waste days hunting through dead MSDN forums from 2004. Instead, we fed raw server logs and screenshots of the errors straight into the AI models. And it was awesome.
We essentially used AI as an interactive, synthetic organizational memory system. The models acted as deep structural translation layers, mapping the archaic execution traits of Windows 2003 directly onto the modern security boundaries, registry virtualization standards, and sub-system changes of Windows 2019. They suggested highly precise work-arounds, identifying exactly which legacy DLLs needed isolation and how to configure our COM+ components to run smoothly without touching a single line of core code. Mostly, the AIs agreed on the technical diagnoses and remedies. But to execute this from the root, we decided to have some fun with the process. Whenever we received a complex remediation plan from one model, we would feed that exact output into the other model as a prompt for a second opinion. Watching two advanced AI models actively debate the finer nuances of 25-year-old Microsoft threading models and memory management wasn't just incredibly entertaining, it forced out edge cases and security vulnerabilities that a human team might have missed entirely. It was peer-review executed at machine speed.
But as technical leaders, we have to look past the immediate wins and ask the hard, foundational question: Can we always trust AI to plug these institutional memory gaps? Is this a bulletproof cure for organizational dementia?
So far, on this complex migration project, my experience has been overwhelmingly positive. I have no doubt that our timeline would have blown out by months, or that we would have been forced into an expensive, high-risk, top-down application rewrite without its helping hand. The AI effectively served as a cognitive bridge over a quarter-century of corporate turnover, which is mind blowing when you think about it!
However, a word of caution to my fellow engineering managers and systems leaders: AI cannot become a total substitute for baseline systems thinking. An LLM understands code syntax, patterns, and error states at a monumental scale, but it completely lacks the implicit business context, the actual human why behind an ancient system design choice. If your engineering team relies blindly on AI-suggested fixes without understanding the underlying mechanics, you aren't actually curing organizational dementia. You are simply replacing old human technical debt with brand-new, synthetic technical debt.
Organizational forgetting is an inevitable law of corporate entropy. Teams will change, documentation will decay, and the bleeding-edge systems you are building today will eventually become the baffling legacy mysteries of tomorrow. But as leaders, our job isn't to mourn the loss of past institutional knowledge, it's to architect resilient, modern methodologies that allow us to adapt when that knowledge vanishes. By leveraging AI to parse our historical codebases, we can reclaim decades of forgotten logic in a matter of seconds, turning technical debt into an addressable, solvable problem.
The next time you find yourself staring at an undocumented legacy codebase, wondering what on earth the original developers were thinking twenty years ago, don't panic. Fire up your LLMs, point them at the root directory, and let machine intelligence help your organization remember what it forgot.
Thanks for listening to Lead Prompt. If today's episode brought some clarity to your technical leadership journey, hit subscribe, share it with your engineering teams, and join me next time as we continue executing leadership from the root. I’m John Collins, and I’ll catch you on the next episode.
References
Remembrance of Things Past? The Dynamics of Organizational Forgetting - https://ideas.repec.org/a/inm/ormnsc/v50y2004i11p1603-1613.html
The Persistence and Transfer of Learning in Industrial Settings - https://ideas.repec.org/a/inm/ormnsc/v36y1990i2p140-154.html
Download audio
File details: 13.7 MB MP3, 10 mins 54 secs duration.
Title music is "Apparent Solution" by Brendon Moeller, licensed via www.epidemicsound.com
Subscribe
Sponsor
Five.Today is a highly-secure personal productivity application designed to help you to manage your priorities more effectively, by focusing on your five most important tasks you need to achieve each day.
Our goal is to help you to keep track of all your tasks, notes and journals in one beautifully simple place, which is highly secure via end-to-end encryption. Visit the URL Five.Today to sign up for free!
