Tech companies keep overlooking the people who know what's broken
The people who understand why systems fail are often the last ones consulted about how to fix them. AI adoption is making that mistake harder to ignore.
Technology companies have a strange relationship with the people who understand how they actually operate.
They celebrate the people who launch things. They promote the people who present strategies. They hire executives to transform processes that nobody bothered to document.
Meanwhile, somewhere in the organization, an employee knows why the Salesforce integration breaks every quarter, why everyone quietly ignores the approval process, and why the company's knowledge base hasn't been trustworthy since 2022.
That person probably isn't presenting at the next leadership offsite.
They're too busy keeping the place running.
The technology industry has confused visibility with value. And AI adoption is exposing just how expensive that confusion can become.
The people who know too much
Every technology organization has people who accumulate institutional knowledge that never makes it into official documentation.
They understand the exceptions. The dependencies. The decisions made three reorganizations ago that still dictate how work gets done.
They know which processes exist for legitimate reasons and which survive because nobody wants to challenge them.
This knowledge is rarely glamorous. It doesn't fit neatly into a quarterly business review or generate impressive LinkedIn announcements.
But remove these people, and organizations often discover how little they understand about their own operations.
The problem isn't simply losing experienced employees. It's losing the operational context that makes documented processes meaningful.
A workflow diagram can tell you what should happen.
The person who maintains that workflow can tell you what actually happens.
Those are two very different forms of knowledge.
Innovation gets rewarded. Operational intelligence gets assigned more work.
Technology companies love builders.
Build a platform. Launch a product. Introduce a new methodology. Announce a transformation initiative.
Creating something new comes with a career narrative.
Why doesn't an existing system work? That's usually considered maintenance.
And maintenance rarely gets the same organizational prestige as innovation.
The irony is that operational intelligence requires many of the same capabilities organizations claim to value in strategic leadership: systems thinking, pattern recognition, risk assessment, and the ability to connect decisions across functions.
Yet employees who demonstrate these capabilities often end up as permanent cleanup crews.
They identify a broken process, so they're asked to fix it.
They recognize a knowledge gap, so they document it.
They understand why a workflow failed, so they inherit responsibility for making it work.
Eventually, their ability to recognize dysfunction becomes an excuse to give them more of it.
Understanding the problem shouldn't automatically make you the unpaid owner of the solution.
AI adoption is making the knowledge problem worse
Now introduce artificial intelligence.
Get Will Kelly’s stories in your inbox
Join Medium for free to get updates from this writer.
Companies want AI agents that automate business processes, retrieve institutional knowledge, and accelerate decisions.
Fine.
But what exactly should these systems learn from?
The outdated SharePoint site?
The Notion workspace nobody governs?
The process documentation written before the last two reorganizations?
Or the Slack conversations where employees explain how work actually gets done?
An AI agent doesn't magically resolve organizational ambiguity. It can automate a poorly understood process just as readily as a well-designed one.
When the underlying knowledge is incomplete, contradictory, or obsolete, automation can spread those weaknesses at machine speed.
This is where the industry's fascination with AI strategy becomes particularly revealing.
Organizations may hire specialists to design sophisticated AI adoption programs without first involving the employees who understand the operational reality those programs must navigate.
The result can be impressive demonstrations built on fragile assumptions.
The AI works beautifully until it encounters the first undocumented exception.
Then the transformation team discovers that the most important information was never in the training materials.
It was sitting with the employee nobody invited to the planning meetings.
Institutional knowledge is infrastructure
We need to stop treating institutional knowledge as an informal employee benefit.
It's operational infrastructure.
That means identifying where critical knowledge lives, assigning ownership, documenting exceptions, and ensuring processes stay understandable when employees change roles or leave.
It also means recognizing that documentation alone isn't knowledge management.
A beautifully organized repository containing outdated information is still an operational liability.
Real knowledge management requires maintenance, accountability, and the authority to challenge processes that no longer make sense.
AI doesn't eliminate these responsibilities.
It makes them harder to ignore.
Stop rewarding the presentation over the outcome
None of this means organizations should stop hiring innovators or pursuing transformation.
They should stop pretending transformation begins with a new tool.
It begins with understanding the business well enough to know what needs changing.
Before launching another AI initiative, organizations should ask who understands the existing workflow, who owns its exceptions, and who will be accountable when automation fails.
Then give those people influence, resources, and recognition, not another cleanup assignment.
The employee who knows where everything breaks isn't necessarily resistant to change.
They may be the person best equipped to make change work.
The future of technology doesn't belong exclusively to the people who can build the next thing. It also belongs to the people who understand why the last thing never worked.
