Can We Really Laugh at NEC's “AI-Only Department”? That Is Not the Real Bottleneck

News that NEC had created an “AI-only department” attracted considerable attention online.

As I scrolled through social media, I saw plenty of cynical reactions:

“They call it AI-only, but humans are still involved.”

“This is why large Japanese corporations are hopeless.”

“They are embarrassingly far behind.”

“They are not using AI. They are just trying to show everyone that they use it.”

I understand the temptation to say these things.

For people who already use AI every day, a company making a grand announcement that it has finally created a department dedicated to AI can prompt the reaction, “Is that really something to advertise now?”

But the news did more than leave me disappointed. It also made me notice something else.

The interesting question is not whether an entire department can operate with AI alone. It is what human work remains after AI is introduced into an existing organization.

When you actually develop software with AI, the answer becomes increasingly obvious.

As AI Gets Faster, Humans Start to Look Slow

AI has significantly accelerated software implementation.

Once the requirements are settled, code implementation, testing, investigation of an existing codebase, and documentation can all proceed remarkably quickly.

The development process as a whole, however, does not accelerate at the same rate.

The moment a decision is required—“Should we choose A or B in this case?” or “How should we handle this exception?”—everything stops.

Humans used to wait for programmers to finish implementing something. Now AI waits for humans to answer its questions.

It is as though the development section of a highway has been expanded to ten lanes, while only one tollbooth remains at the end.

A human is sitting in that tollbooth.

Requirements Work Will Be One of the Last Jobs Standing

Requirements work is particularly difficult.

AI is already quite capable of transcribing a meeting, producing minutes, and turning them into a draft requirements document.

The hard part comes before that.

When you ask people in the workplace what they need, they rarely hand you a clean list of requirements.

Instead, you hear fragments such as:

“This spreadsheet is difficult to use.”

“We have always done it this way.”

“Sales wants this, but accounting will object.”

“Officially, this is prohibited, but this is how we actually do it.”

Someone has to uncover the real problem hidden inside those disconnected comments.

Sometimes the people doing the work cannot clearly articulate the problem themselves.

AI is good at organizing information it has been given. Extracting information that has never been provided by talking to people in the field is a different kind of work.

Human Relationships Make Requirements Harder Than Technology Does

Reconciling competing interests makes the work even more difficult.

Sales wants fewer fields to complete, while accounting wants more. Frontline staff want a simpler process, while administrators need an audit trail. Security wants additional restrictions, while management wants the system delivered quickly and cheaply.

If you give every department’s opinion to an AI, it can organize the options and compare their advantages and disadvantages.

In the end, however, someone still has to say:

“This is what we are going to do.”

That decision may increase somebody’s workload. If the project fails, questions of responsibility will follow.

An AI can generate a hundred proposals. Choosing one, persuading the affected people, and accepting responsibility for the result are separate tasks.

An Ordinary Department Is Harder Than an “AI-Only Department”

That is why my honest reaction to the idea of an AI-only department is: so what?

Introducing AI into an ordinary department is much harder.

As AI takes over research into existing operations, document preparation, implementation, and meeting minutes, humans can concentrate more heavily on judgment and coordination.

Ironically, the more work AI handles, the more the remaining human role becomes a concentrated mixture of listening to stakeholders, reconciling conflicting interests, making the final decision, and taking responsibility for it.

That work does not appear likely to disappear anytime soon.

Then again, an AI-only department does have at least one enormous advantage.

With no humans in the department, meetings to settle the requirements cannot drag on forever.