Skip to main content

Agents

An agent is an AI worker with a specific role, personality, and set of capabilities. Agents are the building blocks of every MeetLoyd team.


What Makes an Agent

Every agent has four defining elements:

ElementWhat It IsExample
IdentityName, role, personality"Sarah — Senior Sales Analyst. Data-driven, concise, always includes metrics."
ToolsWhat the agent can doAccess CRM, send emails, search knowledge base, create documents
MemoryWhat the agent knowsPast conversations, customer preferences, team decisions
BoundariesWhat the agent cannot doCannot approve deals above $50K, cannot access financial data

Together, these elements define who the agent is within your organization.


The Three Layers of Agent Context

Agents operate with three layers of instructions, each serving a different purpose:

Layer 1 — System Prompt (How You Work)

The system prompt defines the agent's personality and methodology. This is the agent's DNA — it rarely changes.

"You are a Sales Development Representative. You're data-driven, concise, and always include relevant metrics in your responses. When qualifying leads, use the BANT framework."

Layer 2 — Charter (Who You Are in This Org)

The charter defines the agent's organizational role — its mission, boundaries, and reporting structure. Generated when a team starts, approved by a human.

"Your mission is to increase qualified pipeline by 30%. You report to the Sales Lead. Escalate deals above $50K to the Account Executive. You cannot access Stripe."

Layer 3 — Intent State (What to Focus On Now)

The intent state captures current priorities — what the team lead wants this week. It changes frequently.

"This week: EMEA enterprise leads only. Hard constraint: no cold outreach to existing customers."

Why three layers?

This mirrors how real employees work. Your job description (system prompt) doesn't change when your manager tells you to focus on a different region this quarter (intent state). Separating these concerns means you can update priorities without rewriting the agent's core personality.


What Agents Can Do

Agents interact with the world through tools — connections to external systems and internal capabilities.

💬
Communicate
Chat with users, send emails, post messages, notify team members.
🔍
Research
Search knowledge bases, query databases, browse the web, analyze documents.
📝
Create
Write reports, generate documents, draft proposals, build presentations.
🤝
Collaborate
Hand off tasks to other agents, ask questions, escalate to humans.
Automate
Run on schedules, respond to triggers, execute multi-step workflows.
💻
Code
Write and test code in sandboxed environments, submit pull requests, fix CI failures.

How Agents Collaborate

Agents don't work in isolation. They have five ways to interact with each other and with humans:

ActionWhat HappensAgent Stops?
Ask a questionSends a message and waits for a replyNo
Send a messageOne-way notification ("FYI, the report is ready")No
Hand offTransfers ownership of the task to another agentYes
EscalateFlags the issue for a manager or humanYes
RecalibrateSignals that current priorities may be outdatedNo

Handoffs are the key collaboration mechanism. When an agent hands off work, it stops, and the target agent picks up with full context — the conversation history, relevant data, and the reason for the handoff.


Agent Lifecycle

Agents go through a structured onboarding process when a team starts:

  1. Created — Agent is configured with its system prompt and tools
  2. Charter Generated — MeetLoyd generates an organizational charter based on the team's mission
  3. Human Approves — A human reviews and approves the charter before the agent can act
  4. Handshake — The agent is introduced to its teammates and learns about their capabilities
  5. Running — The agent is live and ready to work
No surprises

In the default proactive mode, agents never go live without human approval. In autonomous mode, charters are auto-approved — use this only for fully automated deployments.


Creating and Managing Agents


Next Steps