What you'll learn

This section dismantles problematic thinking:

• "SEO agent"
• "Discovery agent"
• "37 agents" architectures

Core lesson:

"Names describe identity. Roles describe authority. Authority must shrink as risk increases.

Bad naming:

• 'Discovery Agent' (names the lifecycle stage - what if it needs to do evaluation too?)
• 'SEO Agent' (names the tactic - what if SEO strategy changes?)
• 'Agent 37' (meaningless - what does it do?)

Good naming:

• 'Search Coordinator' (names the coordination role)
• 'Content Specialist' (names the domain expertise)
• 'Validation Agent' (names the function)

At commit phase, authority shrinks:

• Discovery: One agent can search multiple sources (broad authority)
• Commit: Separate agents for validate, process, confirm (narrow authority each)

Why? Accountability. If something goes wrong during purchase, you need to know:

• Which agent made the decision?
• What data did it have?
• What rules did it follow?

Narrow authority = clear accountability."

This prepares learners to read your diagrams correctly.

Title

Understanding roles inside a multi-agent system

Multi-agent systems only work when roles are clearly defined.
When roles blur, systems become unpredictable, unsafe, and impossible to audit.

This section explains the four most common agent roles used in modern AI products — and how each one affects UX, safety, and system stability.

Why Agent Roles Matter

Humans understand systems through roles:

• Who is responsible?

• Who decides?

• Who executes?

• Who remembers?

Multi-agent systems must reflect this logic.

Without explicit roles:

• multiple agents attempt the same task

• responsibilities overlap

• errors propagate

• recovery becomes impossible

• UX feels chaotic

Clear roles create:
Predictability
Accountability
Safe autonomy
Meaningful control points

Title

What it is

The Orchestrator is the agent responsible for planning, coordination, and delegation.

It does not usually perform tasks itself.

What it does

• Interprets the user’s overall intent
• Breaks goals into sub-tasks
• Decides which agents to involve
• Manages sequencing and dependencies
• Handles retries, fallbacks, and escalation

What it should NOT do

• Call external tools directly (usually)
• Generate detailed content
• Own domain-specific logic

Why it matters for UX

The orchestrator determines:

• What happens first
• What happens next
• When the user is asked for confirmation
• When failures surface

Poor orchestration leads to:

• confusing flows
• silent loops
• missing steps
• unexpected actions

UX rule:
The user should feel guided, not shuffled.