Why Nobody Trains for the Job of Being Human Middleware

Nobody advertised the job. There was no interview for it and no training course. It does not appear on anyone's employment contract or in any performance review. Yet in many businesses, one person does it every day.

They take the orders from the online shop and enter them into the accounting system. They take the payroll export and reshape it so the bank will accept it. They notice when the stock figure in one system no longer matches the other, and they know which one to believe. They are, in the language of software, middleware: the layer that lets two systems that do not speak to each other appear to work together.

Mellorca has written about what it means for a skilled person to spend the day on transfer work in the skilled professional reduced to copy-and-paste. This piece asks a different question: why is this role never named, and what does its invisibility cost the person who holds it?

How the role appears without being created

Consider an illustrative case: a Bloemfontein agricultural supplies business with about forty staff. Six years ago it added an online shop. Three years ago it moved payroll to a cloud service. Last year it began using a delivery-tracking app. Each system was chosen on its own merits and each works well on its own. None of them was connected to the accounting system.

Someone had to close the gaps. The natural candidate was the administrator who already knew the accounting system best. A small task was added: export the shop orders on Monday. Then another: fix the payroll file before upload. Then another: reconcile deliveries with invoices. Nobody decided that she would become the business's integration layer. It happened one reasonable request at a time.

The pattern is described in when your staff become the integration between software platforms and its causes in why high-value employees become human middleware. The human side of it has a particular shape, because the role is never formalised:

It is learned by accident. The person works out the export formats, the quirks and the workarounds alone, often through mistakes.
It is never written down. The steps live in memory and in a few personal notes.
It is not counted. The hours spent do not appear in any measure of the person's work, so they are invisible when workload is discussed.
It is noticed only in absence. When the person is on leave, orders stall and payroll is late, and only then does the business see what she was doing.

What the evidence says about this kind of work

The International Labour Organization's research on generative AI found that clerical work is the occupational group most exposed to automation, with nearly a quarter of clerical tasks highly exposed. Its broader conclusion was that most jobs are more likely to be changed than replaced, and that the larger effect may be on job quality, including work intensity and autonomy. It also emphasised that outcomes depend on management choices, and called for workers' voice and skills training in how technology is introduced.

IBM's explanation of integration platforms describes the problems they exist to solve: applications that multiply, data silos where one part of an organisation cannot see what another holds, and manual data transfer that is slow and error-prone. Microsoft's documentation for its workflow automation service describes creating automated flows between apps and services to synchronise files, collect data and automate repetitive tasks. In other words, connecting systems is recognised as real work with established tools, when it is done by technology. When the same work is done by a person, it rarely receives the same recognition.

The human tax: energy and dignity

Energy is the first cost. Being the link between systems means constant small checks: did the file upload, does this total match, which version is correct. It is a background load that rarely ends, and it often peaks at the worst times, such as month-end or payroll week.

Dignity is the second, and it needs care. Many people take real pride in being the one who keeps things working, and there is genuine skill in it. The concern is narrower. Because the role is unnamed, the skill is unrecognised. It is not on a CV, it is not part of a promotion case, and it is not shared with anyone who could cover. The person may become indispensable and unseen at the same time. For some, that may feel like security. For others it may feel like being trapped in a job that nobody else understands. Mellorca cannot say which applies to any individual.

The business carries a risk too, which Mellorca has described in key-person operational dependency. But this piece is about the person, and for the person the problem is that a significant part of their working life is undescribed.

What could change

The first step is not technology. It is naming the work.

Write the role down. List every transfer, check and reconciliation the person does between systems, with how long each takes. This alone often surprises everyone, including the person doing it. Mellorca's guide on how to find the manual work nobody is measuring sets out one way to do it.
Recognise it in the job. If the connecting work is necessary, it belongs in the job description, workload discussions and reviews.
Share it. A documented process can be learned by a second person, so the first can take leave without anxiety.
Remove the avoidable parts. Many of the transfers can be handled by standard connectors, import templates or simple automated flows. The person's knowledge of the quirks is exactly what is needed to design them well.
Keep the person on the exceptions and the oversight. Automated links still fail. Someone who understands both systems is the right owner for watching them.

Trade-offs

Knowledge can feel devalued. When the connections are automated, the person's hard-won knowledge may seem less necessary. That concern deserves a direct conversation about what the role becomes.
Automation moves the work. Integrations need monitoring and occasional repair. The work changes shape; it does not vanish.
Documentation exposes things. Writing the role down may show how much of it was avoidable, which can feel like criticism of the person rather than of the systems.
Security from indispensability. Some people value being the only one who knows. Sharing the role can feel like losing protection, especially where jobs feel insecure.

Why it matters

The business reasons are clear: fewer errors, less dependence on one person, and capacity that grows without adding staff. The human reason is about being seen. When the connecting work is named, measured and shared, the person who has been doing it gets something back that is easy to overlook: an accurate account of their own job.

That is the dividend. Time, certainly, once the avoidable transfers are automated. But also recognition: the chance for the person to be valued for the judgement they bring to the systems, and to take a week off without the business discovering, too late, everything they quietly held together.

Sources and further reading

Generative AI likely to augment rather than destroy jobs (last verified 10 October 2026)
What Is iPaaS (Integration Platform as a Service)? - IBM (last verified 10 October 2026)
Explore the Power Automate home page - Power Automate (last verified 10 October 2026)