The Day I Thought, “Just Let the AIs Talk to Each Other”—and the Human Work That Remains

Does a Human Really Need to Be in the Middle?

Recently, while I was working, a thought suddenly occurred to me:

“Does a human really need to be involved in this?”

I do not mean that humans are unnecessary.

But I did start to wonder whether people should really be doing the kind of work that amounts to acting as messengers between systems.

The project I am currently working on involves several teams. There is a team responsible for another system, a testing team, and our development team.

What is interesting is that every team now works with AI as a partner.

The testers have AI read the code and write bug reports. Those reports even include detailed explanations such as which line in which file is causing the issue and how it could be fixed.

On our side, we give that Markdown report to an AI and ask it to make the changes. Sometimes we send a request to the team responsible for another system, and they use AI to implement the fix as well.

At this point, the human role is often little more than copying Markdown written by one AI and pasting it into Slack for another.

That made me think:

“Wouldn’t it be enough to connect the AIs through Slack’s MCP integration?”

The testing AI could write the change request. It could pass the request to the development AI through Slack. The development AI could modify the code and create a pull request. A human would review the result at the end and merge it.

That actually sounds like the more natural workflow. At the very least, it would be faster than having me copy and paste Markdown.

I am seriously considering proposing it.

When AI Monitors the Human Monitoring the AI

Thinking about this reminded me of a prediction about the future of aviation that I read years ago.

AI would fly the plane. Takeoff, cruising, and landing would all be automated. So what would the pilot do?

The answer was simple:

“Watch the monitor.”

That alone did not sound too bad. What came next was more unsettling.

To make sure the pilot was actually watching the monitor, the system would track the pilot’s gaze, measure their heart rate, and check whether they had fallen asleep.

In other words, AI would monitor the human whose job was to monitor the AI.

When I first read that, I thought, “Surely that is science fiction.”

Efficiency Is Not the Same as Freedom

These days, however, it no longer seems quite so far-fetched.

If AIs begin passing specifications to one another, writing code together, and reviewing one another’s work, what will be left for humans to do?

Will our job be to press the final approval button? Or to leave evidence that we performed the review properly?

If so, it would not be surprising to receive notifications like these:

• You have not looked at the review screen for the past 15 minutes.
• Your gaze has moved away from the screen.
• Your heart rate is dropping.

I welcome the technology itself. If AIs can communicate directly, that is more efficient. A job that consists of copying Markdown and pasting it into Slack should be one of the first things to disappear.

Still, the history of automation gives me one uneasy feeling.

Humans may be freed from repetitive work. In exchange, we may become the ones being monitored to make sure we are monitoring properly.

It seems that making life more convenient and making people more free are not necessarily the same thing.