I’ve been thinking a lot lately about the way we talk about artificial intelligence.
Specifically, how often we use the term AI when what we’re really talking about is an LLM, or large language model.
That’s not a criticism of LLMs.
They’re incredibly useful tools.
But I think we’re starting to blur the distinction between what these systems actually do and what we assume they can do.
What an LLM Actually Does
At a basic level, an LLM is exceptionally good at working with language.
It has been trained on enormous amounts of information and can use that training to generate remarkably useful responses to questions, summarize information, identify patterns in language, create content, translate, classify information, and perform many other tasks.
That’s powerful.
But there’s an important distinction between generating an answer based on patterns in information and actually analyzing a business problem using operational data and arriving at a reliable conclusion.
Those aren’t necessarily the same thing.
If I give an LLM a description of a warehouse process and ask it to identify potential bottlenecks, it can give me some very interesting ideas.
But if I give it several years of actual warehouse operating data and ask:
Where is the operation losing capacity?
Why is it happening?
What changed?
What should we do about it?
What will happen if we make that change?
That’s a different problem.
And that’s where I think we need to be more careful with the word AI.
The Operations Problem
I’ve spent most of my career in operations and continuous improvement.
One of the things I’ve learned is that before you try to optimize something, you need to understand what is actually happening.
That’s surprisingly difficult.
Companies have systems that don’t talk to each other.
They have spreadsheets sitting between systems.
They have manual workarounds.
They have people who know how to make things work because they’ve been doing it for 15 years.
Eventually, the process works, but nobody can really explain why.
Then someone comes along and says:
“We should automate this with AI.”
Maybe.
But first I want to understand what we’re actually automating.
If the process is broken, AI isn’t necessarily the answer.
It may simply become another layer sitting on top of the problem.
Where I Think LLMs Are Really Valuable
None of this means LLMs don’t have tremendous value in operations.
They do.
In fact, I think we’re probably still underestimating some of their practical applications.
An LLM can be very good at taking information in one form and turning it into something useful in another.
For example:
- Matching information to a known set of requirements
- Summarizing large amounts of information
- Searching and retrieving information
- Creating or modifying documents
- Classifying information
- Helping people interact with complex systems
- Translating information between people, systems, or formats
Those are real productivity gains.
And they’re not trivial.
If I have an employee spending four hours a day moving information from one system into another, and an LLM-based tool can eliminate most of that work, that’s valuable.
But notice what we’re doing.
We’re not necessarily asking the system to think like an operations leader.
We’re asking it to recognize, organize, transform, and communicate information.
And that’s a very different proposition.
The Real Opportunity May Be Between the Systems
This is where I think things get interesting.
In many organizations, the biggest source of inefficiency isn’t inside an individual process.
It’s between processes.
Receiving finishes its work, but the information doesn’t make it cleanly into inventory.
Procurement has one system.
Operations has another.
Customer service has another.
Finance has another.
And somewhere in the middle is Mary, sitting at her desk with three monitors, a spreadsheet, and 15 years of institutional knowledge.
Mary is the integration layer.
She’s making the systems work together because the systems don’t actually work together.
That is an enormous opportunity for technology.
But before we automate Mary, we should understand what Mary is actually doing.
Maybe some of her work should disappear.
Maybe some of it should be redesigned.
Maybe some of it should be automated.
Maybe some of it requires a human decision.
And maybe an LLM is the right tool for part of it.
That’s the distinction I’m interested in.
We Need to Stop Asking “Can AI Do This?”
I think that’s becoming the wrong question.
Almost anything can now be framed as an AI problem.
The better question is:
What exactly are we asking the technology to do?
Are we asking it to retrieve information?
Transform information?
Recognize patterns?
Make a prediction?
Optimize a process?
Make a decision?
Execute a task?
Learn from the outcome?
Those are different capabilities.
And they don’t necessarily require the same technology.
An LLM may be an excellent tool for one and a terrible tool for another.
I’m Not Anti-AI
I’m also not an AI evangelist.
I’m old enough to remember when computers first started entering the workplace.
There were plenty of predictions that computers would eliminate jobs and fundamentally change everything.
They did change everything.
But we eventually learned that a computer is a tool.
It can make people dramatically more productive, but it doesn’t eliminate the need to understand the work.
I think we’re going through something similar now.
The technology is moving incredibly quickly.
The temptation is to start with the technology and then look for problems it can solve.
As an operator, I’d rather start in the other direction.
Start with the problem.
Understand the process.
Understand the people doing the work.
Understand where the friction actually exists.
Then determine whether AI, an LLM, automation, a better process, or simply better management is the right answer.
The technology is the tool.
The problem is still the problem.
