Team Modes

Every team needs a way to organize work. In thinnestAI, the team mode determines how agents communicate, delegate, and produce results. There are five modes:

Mode	Best For	How It Works
Coordinate	Complex multi-step workflows	A leader delegates subtasks to specialists
Route	Request classification	Incoming messages are sent to the best-fit agent
Collaborate	Open-ended problem solving	Agents discuss and build on each other’s work
Tasks	Structured project execution	Work is broken into tasks and assigned to agents
Broadcast	Parallel multi-perspective processing	Input is sent to ALL members simultaneously

Coordinate Mode

A central orchestrator delegates to specialized agents. In Coordinate mode, a leader agent receives the user’s message, breaks it down into subtasks, and delegates each subtask to the most appropriate team member. The leader then synthesizes the results into a final response.

How It Works

User sends a message to the team.
The leader agent analyzes the request and decides which agents to involve.
The leader sends specific instructions to each selected agent.
Each agent completes its assigned work and reports back.
The leader combines the results and responds to the user.

When to Use

Multi-step tasks that require different expertise at each step.
Workflows where order matters (e.g., research first, then write, then review).
When you want a single coherent response that draws from multiple agents.

Example: Research Report Team

# Team: Market Research
# Mode: Coordinate
# Leader: Research Director

# Members:
# - Data Analyst: Pulls market data, creates charts
# - Industry Expert: Provides domain context and trends
# - Writer: Drafts the final report

# User: "Create a market analysis for AI chatbots in healthcare"

# Flow:
# 1. Leader asks Data Analyst for market size data
# 2. Leader asks Industry Expert for healthcare AI trends
# 3. Leader asks Writer to draft report using both inputs
# 4. Leader reviews and delivers final report

Configuration

{
  "team_mode": "coordinate",
  "instructions": "You are a research director. Break down research requests into data gathering, analysis, and writing tasks. Delegate to the appropriate team members and synthesize their work into a comprehensive report.",
  "team_members": [
    {"agent_id": "data_analyst", "description": "Pulls quantitative data and creates visualizations"},
    {"agent_id": "industry_expert", "description": "Provides qualitative industry analysis and trends"},
    {"agent_id": "writer", "description": "Drafts polished reports from raw inputs"}
  ]
}

Route Mode

Routes requests to the best agent based on input. Route mode acts like a smart switchboard. When a message comes in, the team analyzes its content and sends it to the single most appropriate agent. Only one agent handles each request.

How It Works

User sends a message to the team.
The router classifies the message based on agent descriptions and capabilities.
The message is forwarded to the best-matching agent.
That agent handles the request independently.
The agent’s response is returned to the user.

When to Use

Customer support with distinct categories (billing, technical, sales).
Multi-domain chatbots where different knowledge bases apply.
When requests clearly belong to one domain or another.
High-volume scenarios where you want efficient, focused handling.

Example: Customer Support Router

# Team: Customer Support
# Mode: Route
# Leader: Auto-Router

# Members:
# - Billing Agent: Handles invoices, payments, refunds
# - Technical Agent: Troubleshoots product issues
# - Sales Agent: Handles upgrades, demos, pricing questions

# User: "I was charged twice for my subscription"
# → Routed to Billing Agent

# User: "The API keeps returning 500 errors"
# → Routed to Technical Agent

# User: "What's included in the Enterprise plan?"
# → Routed to Sales Agent

Configuration

{
  "team_mode": "route",
  "instructions": "Route customer inquiries to the most appropriate agent. Use agent descriptions to determine the best match.",
  "team_members": [
    {"agent_id": "billing_agent", "description": "Handles billing, payments, invoices, refunds, and subscription changes"},
    {"agent_id": "tech_agent", "description": "Troubleshoots technical issues, API errors, and integration problems"},
    {"agent_id": "sales_agent", "description": "Answers questions about plans, pricing, features, and upgrades"}
  ]
}

Tip: Write detailed, distinct descriptions for each agent. The router uses these descriptions to make classification decisions. Vague or overlapping descriptions lead to misrouting.

Collaborate Mode

Agents work together peer-to-peer. In Collaborate mode, all agents see the conversation and can contribute. There is no central leader — agents build on each other’s responses in rounds. This is useful for brainstorming, review processes, and complex problem-solving where multiple perspectives improve the outcome.

How It Works

User sends a message to the team.
All agents receive the message.
Each agent contributes its perspective or work.
Agents can see and respond to each other’s contributions.
After the configured number of rounds, the final response is compiled.

When to Use

Brainstorming and ideation.
Code review (one agent writes, another reviews, another tests).
Multi-perspective analysis where you want diverse viewpoints.
Creative tasks like content creation with editing.

Example: Content Creation Team

# Team: Blog Post Team
# Mode: Collaborate
# Max Rounds: 3

# Members:
# - Content Writer: Drafts the initial post
# - SEO Specialist: Optimizes for search
# - Editor: Polishes grammar and flow

# Round 1: Writer drafts the post
# Round 2: SEO Specialist suggests keyword improvements
# Round 3: Editor polishes the final version

Configuration

{
  "team_mode": "collaborate",
  "instructions": "Work together to create high-quality content. Each agent should build on the previous contributions.",
  "max_iterations": 3,
  "team_members": [
    {"agent_id": "writer", "description": "Creates initial drafts with engaging content"},
    {"agent_id": "seo_specialist", "description": "Optimizes content for search engines"},
    {"agent_id": "editor", "description": "Polishes grammar, tone, and readability"}
  ]
}

Tip: Set max_iterations to control how many rounds of collaboration occur. More rounds mean better quality but higher latency and token usage.

Tasks Mode

Task-based execution with structured assignment. Tasks mode is the most structured approach. You define specific tasks upfront, and the team assigns each task to the most capable agent. Tasks can have dependencies, priorities, and success criteria.

How It Works

User sends a message or a set of tasks to the team.
The team breaks the work into discrete tasks (or uses predefined tasks).
Each task is assigned to an agent based on capability matching.
Agents execute their tasks, respecting dependency order.
Results from all tasks are compiled into the final response.

When to Use

Project-style work with clear deliverables.
Workflows with dependencies (Task B needs output from Task A).
When you need to track progress on individual subtasks.
Batch processing where multiple independent tasks run in parallel.

Example: Product Launch Preparation

# Team: Product Launch
# Mode: Tasks

# Members:
# - Marketing Agent: Creates launch copy and social posts
# - Technical Agent: Prepares API docs and changelog
# - Design Agent: Creates visual assets and mockups

# Tasks:
# 1. [Technical] Write release notes (no dependencies)
# 2. [Technical] Update API documentation (no dependencies)
# 3. [Marketing] Write announcement blog post (depends on #1)
# 4. [Marketing] Create social media posts (depends on #3)
# 5. [Design] Create launch banner (depends on #3)

Configuration

{
  "team_mode": "tasks",
  "instructions": "Execute product launch tasks in dependency order. Assign each task to the agent best suited for it.",
  "team_members": [
    {"agent_id": "marketing_agent", "description": "Creates marketing copy, blog posts, and social media content"},
    {"agent_id": "technical_agent", "description": "Writes technical documentation, release notes, and changelogs"},
    {"agent_id": "design_agent", "description": "Creates visual assets, banners, and mockups"}
  ]
}

Broadcast Mode

Sends the input to ALL members simultaneously and collects every response. Broadcast mode guarantees that every team member processes the input. Unlike Coordinate (which picks specific members) or Route (which sends to one member), Broadcast fans out to everyone and returns all results.

How It Works

User sends a message to the team.
The message is sent to every team member in parallel.
Each agent processes the message independently.
All responses are collected and returned together.

When to Use

Getting multiple perspectives on a question from different specialists.
Running the same query against different data sources or knowledge bases.
Parallel processing where all outputs are needed (e.g., multi-language translation).
Consensus-building where you want to compare agent opinions.

Example: Multi-Source Research

# Team: Market Intelligence
# Mode: Broadcast

# Members:
# - News Analyst: Searches recent news articles
# - Data Analyst: Queries internal databases
# - Social Analyst: Monitors social media sentiment

# User: "What's the market sentiment on electric vehicles?"

# All three agents run simultaneously:
# - News Analyst: Returns recent EV news coverage
# - Data Analyst: Returns sales data and trends
# - Social Analyst: Returns social media discussion themes

# All responses are collected and returned to the user.

Configuration

{
  "team_mode": "broadcast",
  "instructions": "Send this query to all team members and collect their responses.",
  "team_members": [
    {"agent_id": "news_analyst", "description": "Analyzes recent news coverage and media sentiment"},
    {"agent_id": "data_analyst", "description": "Queries databases for quantitative trends and metrics"},
    {"agent_id": "social_analyst", "description": "Monitors social media for public sentiment and discussions"}
  ]
}

Tip: Broadcast mode is the fastest multi-agent mode since all agents run in parallel. However, token usage scales linearly with the number of members since every member processes the full input.

Choosing the Right Mode

Scenario	Recommended Mode
Customer support chatbot with categories	Route
Research requiring multiple data sources	Coordinate
Content creation with review cycles	Collaborate
Project with clear deliverables and dependencies	Tasks
Getting all perspectives simultaneously	Broadcast
Simple FAQ with one knowledge base	Don’t use a team — a single agent is fine

Combining Modes

You can nest teams. For example:

A Route team at the top level classifies incoming requests.
One of the routed-to “agents” is itself a Coordinate team that handles complex technical issues with multiple specialists.

This gives you the efficiency of routing with the depth of coordination where it matters.

Performance Considerations

Mode	Latency	Token Usage	Complexity
Route	Low (single agent)	Low	Simple
Broadcast	Low (all agents in parallel)	High (all agents run)	Simple
Coordinate	Medium (sequential)	Medium	Moderate
Collaborate	High (multiple rounds)	High	Moderate
Tasks	Variable (depends on dependencies)	Medium-High	Complex

Route is the fastest and cheapest since only one agent runs. Broadcast is fast (parallel execution) but uses more tokens since every member runs. Collaborate is the most expensive due to multiple rounds. Choose based on your quality vs. cost requirements.

​Team Modes

​Coordinate Mode

​How It Works

​When to Use

​Example: Research Report Team

​Configuration

​Route Mode

​How It Works

​When to Use

​Example: Customer Support Router

​Configuration

​Collaborate Mode

​How It Works

​When to Use

​Example: Content Creation Team

​Configuration

​Tasks Mode

​How It Works

​When to Use

​Example: Product Launch Preparation

​Configuration

​Broadcast Mode

​How It Works

​When to Use

​Example: Multi-Source Research

​Configuration

​Choosing the Right Mode

​Combining Modes

​Performance Considerations

Team Modes

Coordinate Mode

How It Works

When to Use

Example: Research Report Team

Configuration

Route Mode

How It Works

When to Use

Example: Customer Support Router

Configuration

Collaborate Mode

How It Works

When to Use

Example: Content Creation Team

Configuration

Tasks Mode

How It Works

When to Use

Example: Product Launch Preparation

Configuration

Broadcast Mode

How It Works

When to Use

Example: Multi-Source Research

Configuration

Choosing the Right Mode

Combining Modes

Performance Considerations