Beyond the hype: AI isn't the box, it's the port

Next week I’ll be in Cartagena, at TOC Americas, talking about artificial intelligence “beyond the hype” to port and terminal operators. And while preparing the talk I realized something uncomfortable: the industry that’s going to be listening has already lived through this exact movie. Not with algorithms, but with a steel box.

The container’s two-decade lag

On April 26, 1956, the Ideal-X sailed from Newark to Houston with 58 containers on board. Today we tell it as the day globalization was born. But in 1956 nobody experienced it that way. The box was a simple idea, almost a dumb one: put the cargo in a standard container and don’t touch it again until it reaches its destination. The box wasn’t the hard part. Everything else was.

For almost two decades the container was, in practical terms, a promise. Ports were designed for longshoremen, not cranes. Unions resisted, and with good reason, because the box made the work they knew disappear. Every shipping line had its own container size, and one line’s ships couldn’t carry another’s boxes. ISO standards weren’t agreed until the late sixties, and it wasn’t until the seventies that the ports rebuilt entirely around the box started leaving behind the ones that simply accepted it on their old docks. Marc Levinson documents it in The Box: the value was never in the container. It was in the system that had to be built for the container to make sense.

The same pattern in every general purpose technology

This isn’t the first time I’ve written about this. In my series on general purpose technologies I’ve traced the same pattern with the ATM, with VisiCalc, with Ford’s assembly line and with the container itself. The technology arrives fast, productivity arrives late, and the gap is always explained by the same thing: how long it takes the organization to rebuild itself around the new tool. Electricity took decades to move factory productivity because owners used it to replace the steam engine in the same spot, instead of redesigning the whole plant. Sounds familiar.

Now swap the box for a language model and the movie repeats itself with uncomfortable fidelity.

What McKinsey’s 2025 data shows

In November 2025 McKinsey published its annual State of AI survey: 88% of organizations already use artificial intelligence in at least one function. Near universal adoption. But only 39% can attribute any impact to their operating results, and most of that 39% say it’s less than 5% of EBIT. Just 6% qualify as what McKinsey calls “high performers,” with real impact on profit. What makes that 6% different? It isn’t better models. According to the same study, what sets them apart is that they redesigned their workflows instead of mounting AI on top of the processes they already had.

In other words, the 6% built the port. Everyone else bought the box and put it on the old dock.

What “beyond the hype” means

This is what I mean by going beyond the hype. It isn’t becoming skeptical of the technology, which works and keeps working better. It’s to stop looking at the box and start looking at the dock. A chatbot that answers questions about a rate changes nothing if the process to approve that rate still takes three emails and a signature. A model that reads shipping documents in seconds is a curiosity if the document still arrives by email to a person who retypes it by hand into another system. I’ve seen it in logistics and I’ve seen it in retail: AI delivered as an isolated tool produces spectacular demos and identical balance sheets.

The problem isn’t technical

And here comes the uncomfortable part, because the problem isn’t technical. The container didn’t need better steel. It needed a port to decide to tear down its warehouses, a union to negotiate a transition and an entire industry to agree on the size of a box. Organizational, political and financial decisions. AI is at the same point. The questions that really matter aren’t which model to use, but which tasks exist only because there was no way to eliminate them before, who reviews when the system gets it wrong, and which area loses power when the process changes. Nobody puts that in the vendor’s deck, and that’s why the vendor sells boxes, not ports.

Substituting vs. redesigning

There’s one distinction I think is central: substituting versus redesigning. Substituting is putting a bot where a person was doing exactly the same task. Redesigning is asking why that task exists. The port that replaced longshoremen with cranes on the same dock gained something. The port that asked why cargo had to pass through a warehouse at all won the decade.

What excites me about saying this at a port conference specifically is that it’s the one industry that can look at the history and recognize itself in it. You know how long it took, how much it hurt and who was left out. And you also know who won: not whoever bought first, but whoever redesigned first.

So the question I want to leave isn’t whether your company is using AI. According to McKinsey, it almost certainly is. The question is a different one: in your organization, which process is still designed for longshoremen, and which one has already been redesigned for the box? And more uncomfortable still: which warehouse does nobody want to tear down?