A 1983 irony that transport software perfected

A 1983 irony that transport software perfected. Bainbridge warned that automating the easy leaves the human alone with the hard. TMS put the human to do machine work. The loose agent threatens to complete the original irony.

In 30 seconds

If in the demo the answer to "what does the human do" is typing, it is the inverted irony. If it is nothing, it is the original irony. If it is deciding the hard with everything served, keep listening.

The 1983 irony

In 1983, British psychologist Lisanne Bainbridge published four pages in Automatica that the technology industry has been relearning the hard way for forty years. The paper is called Ironies of Automation, circulates as a public PDF, and its thesis fits in one sentence. Automating the easy leaves the human in charge only of the hard, precisely when they have already lost the practice to resolve it.

How TMS inverted the irony

Last-generation transport software achieved something notable. It inverted the irony without resolving it. Instead of automating the routine and leaving the hard to the human, it built systems where the human does the machine's routine work. Typing statuses. Capturing rates. Feeding the portal.

The luxury database turned coordinators and providers into capture peripherals, at a measured cost of about $28,500 USD per year per employee involved. The most reliable signal that a TMS failed, sector diagnostics document, is the shadow spreadsheet. The team keeps in Excel what the system should record, because the system demands more than it returns. The machine did not serve the human. The human ended up serving the machine.

How the loose agent completes it

Agentic systems of 2026 threaten to complete the irony in its original version. The agent absorbs routine cases, which are the majority. Only complex exceptions reach the human, which are few but expensive. And if the design treated them as decorative accessories for the compliance slide, they arrive at the decisive moment untrained, without context, and without fresh judgment. The only moment their judgment matters is the moment nobody prepared them for.

The Swiss cheese of escalation

Contemporary supervision literature named the two twin failures that follow. The first is automation complacency. The more reliable the system seems, the less vigilant the supervisor becomes. Outputs are approved without challenge and anomalies are rationalized. The second is teamwork without practice. Handoffs between machine and human become ambiguous and escalation paths are unclear.

These small cracks align like the holes in Swiss cheese, and the error that no single layer would have let through alone crosses all of them at once to payment.

Commercial aviation solved this problem decades ago, with lives on the line. It was not enough to seat a pilot in front of the autopilot and call it supervision. It built Crew Resource Management, recurrent training, and explicit handoff protocols. The lesson is uncomfortable. Presence is not practice. A reviewer without training approving defective outputs is worse than having no checkpoint, because it creates the illusion of one.

Anatomy of a good ticket

This redefines what a good ticket system means. A good escalation is not a tray where rare cases fall. It is a workstation designed with the seriousness of a cockpit.

The ticket arrives with the complete file and evidence already matched, collected where the action happened and not typed by a third party with no incentive. It arrives with the specific reason the agent did not decide, because an escalation without a motive is a guess with a deadline. The reviewer operates with explicit criteria, not inherited intuition. Their reversal rate is measured like any indicator, because whoever approves everything is not reviewing anything. And every human decision returns to the system as training, so the frontier of what can be automated moves each quarter with evidence.

Allocate by capability

Notice the underlying difference with the two failed generations. The luxury database put the human to do machine work and called it digitization. The loose agent put the machine to do human work and called it autonomy. The correct design allocates by capability.

The machine captures, matches, and resolves the repetitive, which is what it does better than anyone. The human judges the exception with evidence served, which is what no honestly calibrated machine can still promise.

In transport operations, where a disputed demurrage can be worth more than the full freight, this distinction is not philosophy. It is the difference between an auditor who signs with a ready file and one who signs because the system looked safe. Bainbridge closed her paper with a warning that aged better than almost all software of her era. The more a system is automated, the more critical the contribution of the remaining human becomes.

At OCL Cargo that allocation is operational: agents on the voyage file; tickets to Finance when money or certainty is insufficient; each correction retrains. It can stamp invoice and Carta Porte.

Sources

• L. Bainbridge, Ironies of Automation, Automatica (1983). Commonly available as an academic PDF.
• Parseur: cost of manual capture
• Galileo: supervision patterns and thresholds

Related reading

• What Decagon understood before the market
• Human escalation is no longer optional
• The war on complexity

FAQ

By Gibrán Ramírez, CEO of OCL Cargo.