What does defining a role actually change?

What material that part carries and what it can reach, which are the two things that most affect the result. That is a real design decision. What it does not change is the underlying capability, so a role is best read as a scope rather than as a qualification.

Why is the planner-and-workers shape so common?

Because it is immediately readable, and at least one framework says that readability is the point. It maps onto how a team would split the job, so the diagram explains itself. That is a good reason to reach for it first and not evidence that it suits any particular piece of work.

Where does this arrangement concentrate its risk?

In whichever part holds the overall goal. Everything below it works inside a plan it produced, so a misreading there is inherited by every piece and corrected by none of them. Each worker can behave impeccably while the whole is pointed somewhere nobody wanted.

Glossary

Agent roles

Agent roles are the parts each participant plays in a multi-part setup, usually one that plans and delegates and others that carry out the pieces.

In plain terms

Deciding who does what before anything starts. One participant plans and hands out work; others take a piece each and report back. The naming is borrowed from how people organise, which makes the arrangement immediately readable and carries an assumption along with it.

Why it matters

Because the vocabulary shapes the design more than anybody intends. Describing a setup in terms of a planner and workers makes that structure the obvious one, and other arrangements have to argue for themselves. The words are doing design work before any decision has consciously been made.

How it works

A role is usually a goal, a set of tools and a description of what this participant is for. That is what the frameworks ask for, and it is genuinely useful: it decides what material each part carries and what it is able to reach, which are the two things that most affect how well any of it works.

The planner-and-workers shape recurs because it is legible rather than because it is optimal. It maps onto how a team would divide the job, so anybody can read the diagram, and the readability is explicitly the pedagogy in at least one framework. Legibility is a real benefit and it is not evidence that the division suits the work.

Naming a role does not create the competence it describes. Calling one participant a reviewer produces something told it is a reviewer, working from the same underlying model as everything else. Where the reviewer and the writer share a blind spot, the review will not find what the writing missed, and the org chart gives no hint of that.

The planning role concentrates the risk, which is the structural consequence people notice last. Whatever holds the overall goal decides what everything else does, so a misreading there propagates through every piece of work below it. The workers can each be reasonable and the whole thing still be pointed in the wrong direction.

What the title says, and what it does

Borrowed vocabulary is doing a lot of work here, and it is worth separating what it delivers from what it implies. Scope, material and reach are real levers: deciding that one part sees only the source documents and another sees only the draft genuinely changes the outcome, and it is the main reason splitting a job helps at all. What the words additionally suggest, without anybody claiming it, is a division of competence. A team that has appointed a reviewer feels it has added a check, and in an organisation of people it would have, because the reviewer brings a different history and different blind spots. Here the reviewer is the same underlying thing reading the same material with a different instruction, so it will tend to find what the writer would have found on a second pass and miss what the writer missed. That is still useful. It is not the independent check the title implies, and the gap between those is where most of the disappointment with these arrangements comes from.

Seen in the wild

A framework where each participant is defined with a role, a goal and tools, then assembled into a crew that collaborates on a task.
CrewAI
A platform where teams of workers are assembled from tools, triggers and instructions to coordinate on bigger jobs.
Relevance AI
An editor whose layer splits a task across parallel helpers, which is the same division without standing job titles.
Cursor

Common misconceptions

People assume

Giving something a role makes it good at that role.

In fact

It tells it what it is for, which changes what it attends to and not what it is capable of. A participant named as a checker is the same underlying model with different instructions, so it shares the blind spots of the one it is checking. The title is a description of intent, not of competence.

People assume

The shape should mirror how a team would divide the work.

In fact

That shape is borrowed because it reads well, not because it fits. People divide work the way they do partly because no one person can hold everything, and that constraint does not transfer cleanly. The familiar diagram is the default rather than the conclusion of any analysis.

Telling them apart

Agent roles vs Multi agent system

Agent roles

What each part is called and what it is for.

Multi agent system

The whole arrangement, including the coordination between parts.

Roles are the labels; the system is what happens between them. Most of the difficulty lives in the second.

Questions

What does defining a role actually change?: What material that part carries and what it can reach, which are the two things that most affect the result. That is a real design decision. What it does not change is the underlying capability, so a role is best read as a scope rather than as a qualification.
Why is the planner-and-workers shape so common?: Because it is immediately readable, and at least one framework says that readability is the point. It maps onto how a team would split the job, so the diagram explains itself. That is a good reason to reach for it first and not evidence that it suits any particular piece of work.
Where does this arrangement concentrate its risk?: In whichever part holds the overall goal. Everything below it works inside a plan it produced, so a misreading there is inherited by every piece and corrected by none of them. Each worker can behave impeccably while the whole is pointed somewhere nobody wanted.

Key takeaways

A role is a goal, some tools and a scope, which decides material and reach.
The planner-and-workers shape recurs because it reads well, not because it fits.
Naming a role does not create the competence it describes.
The part holding the goal concentrates the risk for everything below it.

Tools that use this

CrewAI
Participants defined with a role, a goal and tools, assembled into crews.
Relevance AI
Teams assembled from tools, triggers and instructions.
Cursor
Parallel helpers, the same division without standing job titles.