Everyone is talking about AI.

Every software company has an AI roadmap. Every conference has AI panels. Every executive meeting eventually finds its way back to AI. In many ways, that's understandable. The technology is advancing quickly, and there are legitimate opportunities to improve productivity, automate repetitive work, and help people make better decisions.

But the more conversations I have, the more I notice a recurring pattern.

A company has a real operational problem.

Quoting takes too long.
Inventory can't be trusted.
Teams are duplicating work.
Information is spread across multiple systems.
Employees spend hours every week trying to determine which version of the truth is actually correct.

Then someone says, "We need AI."

Suddenly the discussion shifts away from fixing the operational roadblocks and toward deploying intelligence.

New agents get introduced. New automation layers appear. Dashboards get smarter. More technology gets added.

Yet underneath all of it, the operating environment remains exactly the same.

The workflows are still fragmented while approvals still happen outside the system because the data still exists in multiple places. And everyone is still manually reconciling contradictions.

That's why I keep coming back to the same metaphor.
​

​
Sure the tools changed. But the problem didn't.

---

Strong Employees Often Hide Weak Systems

One of the biggest misconceptions in business technology is the assumption that digitization automatically creates operational maturity.

It doesn't.

A company can have software everywhere and still be very heavily dependent on human interpretation.

In fact, many cpompanies are running on what I would call human stitched architecture. The process technically exists inside the system, but the process only works because someone with experience constantly compensate for its weaknesses.

• The inventory manager knows which quantity can't be trusted yet.
• The sales team knows which spreadsheet contains the real pricing.
• The buyer knows which supplier status is wrong.
• The warehouse supervisor knows which workflow everyone bypasses when things get busy.

The software says one thing and reality says another so experienced employees become the bridge between the two.

What's interesting is that these people often make the company look healthier than it really is. Customers get served. Orders get shipped. Revenue grows. Leadership sees results.

But behind the scenes, people are constantly correcting the system.

I've seen plenty where the most valuable operational person wasn't the CEO, the VP of Operations, or the ERP administrator.

It was the employee who knew where all the contradictions lived.

Every company has a version of that person and that person is often carrying more operational risk than anyone realizes.

---

The Problem With Human Glue

The challenge is that we slowly adapt to these conditions and over time, the workaround becomes the process. Approvals start happening through conversations instead of workflows. Exceptions become normal. Departments create their own tracking sheets. And all of a sudden seemingly, many maintain their own versions of information because trust in the central system has eroded.

The business keeps moving, so nobody feels immediate pressure to address it. In fact, many companies can operate like this for a while. Some become highly successful while doing it.

But success doesn't eliminate operational debt.

It often hides it.

Every spreadsheet introduces another uncontrolled state every manual reconciliation creates another opportunity for divergence, and every workaround creates another dependency on a specific person.

The debt remains mostly invisible until something exposes it.

A key employee leaves.

An audit occurs.

The company doubles in size.

Or increasingly, automation gets introduced.

---

AI Doesn't Create Clarity

One of the reasons I think AI conversations sometimes go sideways is because people assume AI automatically creates intelligence inside an organization.

I don't think that's actually what happens.

What AI does is introduce acceleration.

Whether that acceleration creates value depends entirely on the quality of the environment underneath it. Why?

Because fragmented workflow executed faster is still in pieces. A disconnected approval chain automated through agents is still disconnected. Or a recommendation engine operating on conflicting information still produces recommendations based on conflicting information.

The technology itself doesn't resolve ambiguity. It scales whatever already exists.

That's why I think operational architecture deserves far more attention than it currently receives. Because architecture determines what intelligence has access to.

And more importantly, whether the information being used is trustworthy in the first place.

---

Four Different Truths

I was speaking recently with a company that believed they had strong inventory controls because every department technically used the same ERP.

From a distance, everything looked centralized. But as we dug deeper, a different picture emerged.

The team that quotes had and maintained its own spreadsheet because they didn't fully trust inventory visibility. At the same time, purchasing maintained separate tracking because supplier updates often lagged reality. And the warehouse occasionally skipped receiving steps during urgent situations. And don't forget about finance, which regularly compensated for recurring discrepancies downstream.

Nobody was doing anything malicious. Everyone was simply trying to keep the operation moving. Yet four different versions of truth existed simultaneously.

This is where I think many AI conversations become uncomfortable.

• Which state is authoritative?
• Which dataset becomes the foundation for future recommendations?
• They're operational design questions.

And until they're answered, intelligence is operating on uncertainty.

---

The Real Issue Is Operational Ambiguity

Most operational problems eventually lead back to the same root cause. The system itself doesn't fully understand operational reality and a quote gets approved against outdated inventory.

A shipment moves before documentation is complete while a repair advances while certification issues remain unresolved, and an approval happens in email while the ERP still shows pending status.

Each event by itself seems manageable, but over time they create a growing gap between official workflows and actual workflows.

That gap matters.

Especially once automation enters the picture, because intelligence performs best when context exists, such as:

1. Reliable event history.
2. Enforced workflows.
3. Intentional exception handling.
4. Trustworthy records.

Without those foundations, intelligence becomes speculative instead of contextual. And speculative systems become increasingly difficult to govern as they scale.

---

Why Closed-Loop Systems Matter

The companies that benefit most from AI over the next decade won't necessarily be the companies deploying the most agents.

I believe they'll be the companies that invest in operational coherence first.

That's where closed-loop environments become important.

In a closed-loop system, actions, approvals, permissions, events, and historical state remain connected. The system understands not only what happened, but the conditions surrounding what happened.

That continuity creates context which is ultimately what makes intelligence useful.

The industry spends a lot of time discussing model capability.

I think context deserves equal attention.

A moderately intelligent system operating inside strong operational flow will often outperform highly advanced intelligence operating inside poor or broken architecture.

The same thing is true for people.

Good operators make better decisions when workflows are clear, responsibilities are defined, and operational history can be trusted.

Strong architecture improves decision-making itself.

---

Building Intelligence Inside Structure

When we began thinking about ELIA, we weren't interested in adding AI for the sake of adding AI. The more difficult challenge was determining how intelligence could operate safely inside live operational environments, and that required a different approach.

Intelligence needed context. It needed awareness of workflow state, permissions, event history, and transactional relationships.

Recommendations needed to inherit context.

Actions needed to inherit governance.

Exceptions needed to inherit routing.

In other words, intelligence needed operational boundaries.

Because intelligence without structure eventually becomes another source of ambiguity.

---

The Companies That Win Will Think Differently

Eventually, powerful AI will be widely available.

When that happens, access to models won't be a meaningful differentiator.

Operational coherence will.

The companies that succeed won't necessarily be the ones talking about AI the most.

They'll be the companies whose workflows, governance structures, permissions, traceability, and operational continuity are strong enough to support intelligence without creating instability.

Because underneath most operational failures isn't a lack of intelligence.

It's ambiguity, about:

• Ownership
• Process and state
• and about which system actually represents truth.

AI doesn't automatically remove that ambiguity.

Sometimes it amplifies it.

Which is why I believe organizations should spend less time asking how intelligent the technology is becoming and more time asking whether the environment around that technology is coherent enough to support it.

Otherwise, we're simply replacing one tool with another sharper one and mistaking movement for progress.

The skewer still isn't the spoon.