The Jobs That Don’t Have Names Yet

Most discussions about artificial intelligence and jobs begin with the same question: which jobs are going to disappear?

It is an important question, but I am beginning to think it may not be the most interesting one.

A hundred years ago, nobody was training to become a cybersecurity analyst, an app developer, a cloud architect or a social-media manager. Those jobs did not exist because the problems they solve did not exist.

So perhaps the better question is not simply which jobs AI will eliminate. It is what new problems AI, robotics and the next generation of technology are creating that somebody is eventually going to get paid to solve.

We probably cannot predict the job titles of 2040. But we can already see some of the work.

Somebody Has to Supervise the Agents

One of the easiest future roles to imagine is the person supervising groups of AI agents.

If one employee eventually manages five, twenty or fifty agents, somebody has to assign objectives, watch results, catch mistakes, resolve conflicts and decide when human intervention is necessary. But that is only part of the job.

The supervisor also has to know whether all that activity is actually moving the organization toward its larger objectives.

That means agent supervision cannot simply become a technical function isolated somewhere inside IT. It has to connect upward into enterprise goals, operating priorities and management decisions.

A group of agents could be enormously productive while still heading in the wrong direction.

We have already seen versions of that problem. More output does not necessarily mean more progress.

The future agent supervisor may therefore look less like a programmer and more like a manager with an unusual workforce. One question will be whether the agents are doing what they were told. The more important question will be whether they are doing what the organization actually needs.

Those are not the same thing.

Someone Has to Write the Script

There will also be people whose work is translating human intent into instructions machines can actually follow.

That sounds a little like prompt writing today, but I suspect it becomes much broader. The closest analogy may be a script writer.

The players are the AI agents. Someone has to define the dialogue, the rules, the sequence, the boundaries and the amount of acceptable improvisation.

Unlike a movie script, however, this one keeps changing because the business keeps changing.

Tell an agent to handle customer complaints and you immediately discover that the instruction is not nearly specific enough. What counts as a complaint? How much money can the agent refund? When does it escalate? What happens if the customer threatens legal action? What if the customer is clearly wrong but also extremely valuable? What if company policy conflicts with what makes business sense?

That requires more than technical instructions. It requires someone who understands the business well enough to build the operating logic the agents will use.

Today we might call that person a process designer, systems analyst, operations manager or something else. In the future, the responsibility may become specialized enough to acquire a name of its own.

Robots Will Need Mechanics

Some future work is easier to visualize because we have seen the pattern before.

The automobile created the automobile mechanic.

Robots will create something similar.

A warehouse robot, delivery robot, hospital robot, construction robot, household robot or robot gardener will eventually break, get stuck, lose calibration, damage something or encounter a situation it cannot resolve.

Someone will have to show up with diagnostic equipment and tools.

That person may need to understand software, sensors, motors, batteries, communications and mechanical systems. The robot mechanic may look less like the mechanic of 1960 and more like a hybrid of electrician, computer technician and equipment specialist.

The title may change. The underlying need will not.

Machines that do useful work eventually need people who know how to keep them doing it.

AI Still Needs a Physical World

Artificial intelligence is often described as if it lives somewhere in the cloud.

The cloud, unfortunately, has buildings.

Very large buildings.

Those buildings need electricity, cooling, networking, backup power, security, maintenance and repair. They need electricians, HVAC technicians, grid specialists, fiber technicians, equipment installers and people who understand increasingly complicated power and cooling systems.

This is where the future of work becomes slightly ironic.

AI may automate some forms of office work while increasing demand for people who install, maintain and repair the physical infrastructure that makes AI possible.

Many of those people may never think of themselves as working in artificial intelligence.

They will simply be doing jobs that exist because AI does.

The People Nobody Notices Until Everything Breaks

There is another category that already exists inside many organizations.

These are the people who understand how all the systems actually fit together.

They may be buried several levels down in the organization. They know which old application still feeds data to the new one. They know the workaround nobody documented. They know why two systems that supposedly integrate do not quite integrate, and they know what happens when one process fails and three unrelated departments suddenly stop working.

They are frequently undervalued because, most of the time, nothing dramatic happens.

That is partly because they are doing their jobs.

Then something breaks and nobody else knows the answer.

Some upper-middle manager usually remembers that there is one person somewhere in the organization who understands how the whole mess actually works.

Call her. Or him.

AI may make these people considerably more important. Organizations are going to spend years operating mixtures of legacy systems, cloud applications, human processes, AI agents and automated equipment. The interface problems will multiply, the integration problems will multiply, and the number of things that can fail in unexpected combinations will multiply.

The people who understand those connections may become some of the most valuable employees in the building.

Whether anyone notices before something breaks is another question.

When Something Goes Wrong, Who Figures Out Why?

AI also creates a new version of an old profession.

Auditors, investigators and compliance people already exist. But what happens when an autonomous system makes the wrong decision?

Why did the insurance system deny the claim? Why did the purchasing agent order 10,000 of the wrong component? Why did the scheduling system cancel the wrong appointment? Why did the factory optimization software increase output while quietly reducing quality?

Someone will have to reconstruct what happened.

In many cases, the answer may not be that the machine malfunctioned. It may have followed its instructions perfectly.

The mistake may have been in the objective, the data, the assumptions or the boundaries.

That makes the investigator’s job part technical, part operational and part managerial.

And as AI systems are given more authority, that work becomes more important because the cost of a bad assumption can spread much faster.

What Does Authentic Mean When AI Runs the Process?

Authenticity seems relatively easy to define in media.

Is this photograph real? Was this video generated? Did this person actually write that article? Was that voice recording authentic?

Those questions will create plenty of work in journalism, entertainment, cybersecurity and fraud prevention.

But the more interesting problem may appear when AI begins controlling essential processes.

Suppose AI manages part of an assembly line. What does “authentic” mean there?

Is the production report authentic because the machines actually produced the numbers? What if the AI changed the operating process in a way nobody specifically approved? What if the reported output is technically accurate but hides increasing defect rates? What if the automated system optimizes production so effectively that managers no longer fully understand how the result was achieved?

At that point, authenticity may no longer mean simply, “Was this created by a human?”

It may mean whether we can verify what actually happened, reproduce it, explain why the system made the decisions it made and identify who is responsible for the result.

That is a much broader problem than detecting fake photographs.

We may need people whose job is not simply proving that information is real, but proving that automated processes are understandable, traceable and trustworthy.

Management Does Not Disappear

There is one category I initially thought might become a new job: the people who decide what machines should do.

The more I think about it, the less new that sounds.

That is management.

Managers and executives are already supposed to set objectives, establish priorities, make judgments, allocate resources and change direction when circumstances change. AI does not eliminate those responsibilities.

It may make them more important.

If machines become extraordinarily good at execution, then mistakes in direction become much more expensive because the organization can execute the wrong strategy faster than ever.

For a long time, much of a worker’s value came from knowing how to do something. AI may make some kinds of “how” increasingly cheap.

That makes “what,” “why” and “whether” more valuable.

What should we be doing? Why are we doing it? Should we still be doing it at all?

Those are old management questions.

The technology simply makes bad answers more dangerous.

But Who Trains the Managers?

That creates an immediate problem.

If existing managers are suddenly responsible for supervising AI agents, evaluating automated decisions and managing organizations where humans and machines share the work, where do they learn how to do it?

The answer probably comes from both directions.

From the top down, companies will need formal training, standards and outside expertise. Some of that will come from the large technology companies building the systems. Some will come from consulting firms, systems integrators, universities and training organizations. Experienced people will undoubtedly move from technology companies into traditional enterprises, either as employees, advisers or implementation specialists.

They can teach organizations what the technology can do.

But that will not be enough.

The people arriving from the technology side may understand the tools much better than they understand the business they are being inserted into.

That is where the bottom-up learning becomes essential.

Experienced employees already inside the organization know the exceptions, the workarounds, the strange dependencies and the situations where judgment matters more than procedure. They are the people likely to ask questions management may not yet know to ask.

What happens when this case does not fit the model?

Who is accountable if the agent makes this decision?

What information does the AI not know?

What happens if two automated systems optimize for different objectives?

Are we solving the right problem?

Some of those questions may initially sound like resistance to change.

Often, they will actually provide the restraint and balance the organization needs.

The technology leaders can teach what the systems can do.

The experienced operators can teach where they are likely to fail.

Management has to make those two forms of knowledge meet.

The Jobs Will Probably Appear Before the Titles Do

That may be how many of these future jobs actually emerge.

They will not begin with somebody in human resources inventing a clever title with a job description and then hiring for it.

The work will appear first.

• A department manager gradually discovers that half the workforce being managed is software.
• A compliance person starts getting all the strange AI-related cases.
• A maintenance technician becomes the person everybody calls when the robots behave oddly.
• A systems analyst becomes indispensable because nobody else understands how the old applications, new cloud systems and AI agents connect.

At first, these are simply additional responsibilities added to existing jobs.

Then the responsibilities become large enough that somebody recognizes them as a specialty.

Training follows.

Standards develop.

Eventually, somebody gives the work a title.

The progression may be something like this:

The problem appears.

• Someone figures out how to handle it.
• Other people begin doing the same work.
• Training develops.
• Standards emerge.
• And finally the job gets a name.

That is probably more realistic than trying to publish a list today of the ten hottest careers of 2040.

We Have Seen This Before

The automobile did not simply eliminate horse-related work.

It created mechanics, filling stations, dealerships, traffic engineers, highway construction, insurance adjusters, driving instructors, parking garages and industries nobody could have fully predicted when the first cars appeared.

The Internet did the same thing.

Thirty years ago, very few parents were telling their children to become search-engine optimization specialists, cloud architects, social-media managers, app developers or cybersecurity analysts.

Not because those were bad career choices.

The jobs did not have names because the problems did not exist yet.

AI and robotics will probably follow the same pattern. Some jobs will disappear. Some will shrink. Some existing jobs will become more important. Entirely new kinds of work will emerge around problems we have only begun to encounter.

The Jobs May Follow the Problems

Trying to predict exact job titles twenty years from now may be mostly entertainment.

But that does not mean we know nothing.

We can already watch where the problems are accumulating, where responsibilities are expanding and where organizations keep discovering work that somebody has to own.

Near term, much of that work will probably be absorbed by people who already have jobs.

In the interim, repeated problems will create specialties, training programs and formal responsibilities.

Longer term, some of those specialties may become occupations we do not yet have names for.

That is not much different from previous technology revolutions. The technology arrives first. The problems follow. People begin solving them. Eventually the labor market catches up.

The jobs of the future may not be defined simply by what humans can do better than machines.

They may be defined by the new problems the machines create.

And if history is any guide, there will be plenty of those.