When Employees Become the Architecture

There is a particular kind of technology problem that interests me because it can remain almost invisible for years.

The system works.

Nothing is technically broken.

The records save correctly. The automation runs. The integration sends information where it is supposed to go. The organization continues operating.

And yet everybody who actually uses the system has developed a private collection of knowledge required to make it work.

Someone knows that one field should never be trusted.

Someone else knows that a particular report has to be exported and corrected manually.

Another employee knows that when the system says one thing, it actually means another.

A project manager remembers why an automation was created because the original decision was never documented.

An administrator knows which process exists for a legitimate reason and which one exists because somebody needed a temporary workaround five years ago.

The technology is functioning because the people around it have learned how to compensate for it.

I think this distinction matters.

We usually describe technical debt in relation to software. A compromise is made today that creates additional technical work later.

But organizations can accumulate another kind of debt.

I think of it as cognitive debt.

Cognitive debt develops when the architecture repeatedly requires human beings to remember information the system should preserve, interpret distinctions the system should make explicit or manually reconcile relationships the system should already understand.

One workaround is manageable.

Hundreds of small workarounds eventually become an organizational culture.

This is particularly interesting because human beings are extremely adaptable.

Give someone an inefficient system for long enough and she will develop an efficient way of surviving it.

Then something strange happens.

Her adaptation begins to conceal the original failure.

The employee becomes part of the architecture.

I think this is one of the reasons organizations can underestimate how much their technology is costing them.

The cost does not always appear as downtime.

It appears as thought.

It appears in the ten minutes somebody spends searching for information that technically already exists.

It appears when an employee has to ask the one person who understands why something works the way it does.

It appears when a new employee requires months to learn rules that are nowhere in the documentation.

It appears when two departments manually reconcile information because their systems represent the same concept differently.

Individually, these moments seem trivial.

Collectively, they represent an enormous amount of human intelligence being spent compensating for architectural ambiguity.

This becomes especially important as companies begin introducing AI.

There is a temptation to place an intelligent layer over an unintelligible system.

I understand why.

If employees struggle to find information, give them an AI assistant that finds it.

If documentation is disorganized, let an LLM interpret it.

If processes are complicated, place a conversational interface in front of them.

Some of these applications are genuinely useful.

But I think there is a danger here.

AI can reduce cognitive debt.

It can also hide it.

If an AI system becomes responsible for translating an incoherent information environment into something a human being can understand, the underlying incoherence has not necessarily disappeared.

We have introduced another interpreter.

This is why I think AI implementation should force organizations to examine their architecture more carefully rather than allowing them to examine it less.

Before asking what an AI agent should do, I want to know why the process requires an agent in the first place.

Before asking a model to reconcile conflicting information, I want to know why the information conflicts.

Before asking AI to explain a system to an employee, I want to know why the system became so difficult to understand.

Sometimes AI will still be the correct answer.

But now we understand the problem it is solving.

That distinction matters enormously to me.

At North Shore, I do not think the purpose of technical consulting should be to make the client increasingly dependent upon people who understand complicated technology.

I think the opposite is more interesting.

The system should become increasingly understandable.

Architecture should preserve its own reasoning. Documentation should preserve decisions. Information should have an identifiable source. Processes should make sense to the people responsible for them.

The consultant should absorb complexity rather than transfer it to the client.

A sophisticated system does not need to feel sophisticated to use.

In fact, I think the opposite is often true.

When technology has been considered deeply enough, the person using it should encounter less complexity than the people who designed it.

That may be one of the best measures of good architecture.

The complexity still exists.

It simply no longer belongs to the user.

Previous
Previous

AI at Work: A Useful Tool Paired with Human Judgment

Next
Next

I Think We Have Removed Too Much Art From Technology