There is a serious debate underway about what AI will do to employment.

Some of the evidence is reassuring. New technology can make workers more productive, lower costs, increase demand and create kinds of work that didn't previously exist.

Some of it isn't.

Entry-level work is especially interesting because those jobs often do more than produce useful output. They're how inexperienced people become experienced people. Remove enough of that work and we may discover that we've automated part of the process by which the next generation learns how to do the jobs we still need humans to do.

I don't know how that resolves at the scale of an entire economy.

I do know what the question looks like inside a small newspaper.

We have journalists who have spent decades learning their communities.

They know which meeting is likely to matter before the meeting happens. They recognize a name in a document. They remember the earlier version of an argument that has suddenly become new again. Someone calls them because a relationship was established years ago, often through circumstances nobody thought to record in a database.

That's work.

So is taking the article that reporter wrote in Word, formatting it for the website, identifying the photograph, entering an event into a calendar, creating metadata and moving information from one system into another.

They're not the same kind of work.

For years, small organizations have dealt with that distinction by asking people to do both.

It happens gradually.

Someone keeps a spreadsheet because there isn't a proper system. Someone remembers which advertiser needs to be called because the software doesn't. An email inbox becomes a workflow. A manager checks three places every morning because none of them talks to the others.

Nothing is completely broken.

That's why it can continue for years.

The organization adapts by spending human attention.

Usually the most conscientious people absorb the difference. They remember. They double-check. They stay late. They develop little personal systems that keep everything moving.

Eventually their improvisation becomes part of the infrastructure.

Except it isn't infrastructure.

It's a person compensating for the absence of it.

That distinction has become much more important to me as AI has improved.

Our reporters don't need to become web developers.

If someone has spent thirty years learning how to report on a county, teaching them to care about HTML structure would be a strange use of that experience.

They can write the article.

Software can help translate it into the forms needed elsewhere.

That may include identifying names and places, suggesting structure, preparing a web version or recognizing that something in the story belongs on the community calendar. A person can review the result and decide whether it should be published.

AI removes some work.

That's the point.

The interesting question is which work.

If we automate the part where a reporter talks with people, attends meetings, recognizes changes, asks uncomfortable questions and decides what matters, we've changed the newspaper considerably.

If we automate copying information from one form into another, we've changed something else.

We've given some attention back to the reporter.

The boundary won't always be that clean.

AI will become capable of things we currently regard as requiring human judgment. People will also continue doing some routine tasks simply because automation isn't always worth the trouble.

So I don't think “humans do judgment and machines do repetition” is a permanent rule.

It's a useful question to keep asking.

Where does the knowledge come from?

Who experiences the consequences of getting something wrong?

What part of the work develops capabilities we'll need later?

What part exists mostly because two systems don't communicate?

And what happens to the person when that task disappears?

That last question matters.

A junior reporter may spend time doing routine work while learning how the newspaper operates. Automating all of it could free them to learn faster.

It could also remove the path by which they used to learn.

We won't know which merely by measuring productivity.

This is where the national conversation can become difficult to recognize from inside a small institution.

At large scale, a job is necessarily a category.

Locally, it's usually a person.

You know who does it.

You know what else they know.

You may discover that the apparently inefficient employee is carrying three relationships, two undocumented procedures and fifteen years of institutional memory that weren't included in the job description.

Automation looks different once you can see that.

The goal can't simply be to identify tasks a machine can perform and remove the humans performing them.

Capability isn't the same as wisdom about where to use it.

We've been learning that while rebuilding the digital infrastructure around the Mountain Eagle.

The visible part is a newspaper website.

Underneath it are subscriptions, advertising, directories, events, publishing tools, archives, payments and other ordinary things a regional newspaper needs to function.

None of those things is particularly exciting by itself.

Together they can remove a surprising amount of work that previously had to live somewhere in a person's day.

That's what I increasingly want infrastructure to do.

Not make people unnecessary.

Make unnecessary demands on people unnecessary.

There is an important difference.

It also changes how I think about owning the infrastructure.

If more of an organization's work is being mediated by software, then whoever controls that software occupies a more important position than they did before.

That doesn't mean every organization should own every piece of technology it uses. We use outside services too. Sometimes renting something is considerably more sensible than building it.

The question is which relationships become dangerous to lose.

A newspaper can change payment processors.

Losing its archive would be something else.

It can use social media to reach readers.

Allowing a social network to become the only place it knows those readers would create a different dependency.

As AI becomes another layer in this arrangement, the same distinction applies.

Use it where it increases capability.

Be careful when convenience quietly becomes dependency.

And keep enough understanding close to the people experiencing the consequences that someone can still notice when the system is doing the wrong thing very efficiently.

This doesn't answer whether AI will create or destroy more jobs nationally.

I'm not sure anybody can answer that yet.

It does suggest that employment totals aren't the only thing worth watching.

A community can have jobs and still lose capabilities.

An organization can become more productive while becoming more fragile.

A worker can be relieved of tedious work and become considerably more valuable.

Another worker can lose the routine work through which they would have learned something important.

All of those things can happen at the same time.

Maybe that's one advantage of looking at the problem from somewhere small enough to know the people involved.

The categories start breaking apart.

You stop asking only whether AI can do the work.

You start asking what the work was doing for us.