Over the past few years the AI ecosystem has been assembling itself into layers.

First came the models. Then came the tools that allow those models to interact with the world. Now we’re beginning to see protocols that let AI agents communicate with each other and frameworks that help orchestrate their work.

But when you zoom out and look at the emerging architecture, a small question starts to nag.

What is the unit of work in AI systems?

Not a prompt.

Not a tool call.

Not a message between agents.

Something more like what humans already understand: a mission.

In this episode we explore a simple but surprisingly deep idea: that AI systems may eventually need a shared way to describe purposeful work — goals, constraints, policies, and budgets — independent of the particular agents or tools involved.

Along the way we talk about:

Why the AI stack may be missing a coordination layer

The difference between agents, tools, and missions

Why reasoning and authority should probably be separated

How runaway agent systems could create congestion

Why TCP solved packet congestion — but not “work congestion”

What might stop agents from spawning missions all the way down

Whether this is just reinventing workflow systems

And why the hardest problem in large systems is often coordination, not intelligence

The conversation is exploratory rather than prescriptive. The point isn’t to propose a standard — at least not yet — but to ask whether the ecosystem might be approaching the kind of scale where coordination layers historically appear.

Because once AI systems start generating work for each other, the central question changes.

Not what can these systems do?

But how many of them can operate together without overwhelming the environment they share?

Get full access to The Generative Gazette at generativegazette.substack.com/subscribe