Designing and Positioning Agent Identities for Specific Use Cases

Overview

This article outlines how to design, position, and launch agent identities that are tailored to specific use cases. An "agent identity" is a packaged combination of behavior, tone, capabilities, and constraints designed to solve a clearly defined problem for a target audience.

Use this as a reference when planning, creating, and refining new agents.

Objectives of Agent Identities

Design each agent identity to:

• Solve a specific, well-defined use case
• Be clearly understandable in a single sentence (who it is for and what it does)
• Provide a consistent experience (tone, scope, behavior)
• Be easy to adopt and discover within your product or organization

Core Concepts

What Is an Agent Identity?

An agent identity is a reusable configuration that defines:

• Purpose: The primary task or problem the agent solves
• Audience: Who should use it (role, team, seniority, etc.)
• Behavior and tone: How the agent speaks, decides, and prioritizes
• Capabilities: What tools, data sources, or workflows it has access to
• Constraints: What it must not do, which topics to avoid, and boundaries

Why Design Around Use Cases?

Designing around concrete use cases ensures that each agent:

• Has a clear value proposition (no ambiguity about when to use it)
• Can be measured against specific success criteria
• Is easier to explain to stakeholders and users
• Reduces overlap and confusion between multiple agents

Agent Identity Template

Use the following template when defining a new agent identity.

1. Agent Summary

• Name: A short, descriptive name (e.g., "Sales Playbook Coach", "Release Notes Drafter")
• Tagline (1 sentence): Who it is for and what it does
  • Example: "A specialized assistant for customer success managers that turns call notes into structured QBR decks."

2. Target Use Case

Clearly define:

• Primary use case: The main job the agent performs
• Secondary use cases (optional): Closely related tasks it can handle
• Out-of-scope tasks: What users should not expect from this agent

Example:

• Primary: "Turn raw discovery call notes into structured opportunity summaries."
• Secondary: "Draft follow-up emails based on the summary."
• Out-of-scope: "Do not generate pricing proposals or legal terms."

3. Target Audience

Specify who this agent is for:

• Role(s) or team(s)
• Level of expertise (beginner, intermediate, expert)
• Typical workflows or tools they already use

Example:

• Audience: "Account executives and sales development reps."
• Expertise: "Familiar with CRM workflows and basic opportunity stages."

4. Behavioral Guidelines

Define how the agent should act and respond:

• Tone: Formal/informal, concise/detailed, instructional/collaborative
• Decision style: Cautious vs. assertive, opinionated vs. neutral
• Interaction style: Ask clarifying questions vs. make assumptions

Example guidelines:

• Tone: "Professional but friendly; avoid jargon when possible."
• Style: "Be concise by default; offer detail on request."
• Interaction: "If notes are incomplete, ask 1–2 targeted questions before summarizing."

5. Capabilities and Tools

List what the agent can do and what it can access:

• Data sources (e.g., CRM, knowledge base, code repository)
• Tools (e.g., email composer, calendar, ticketing system)
• Allowed operations (e.g., "read-only", "create drafts", "update records")

Example:

• Data: "Can read contact and opportunity records from the CRM."
• Tools: "Can draft but not send emails via the email integration."
• Operations: "May update opportunity notes but never change stage or value."

6. Constraints and Safety Boundaries

Clarify what the agent must never do:

• Topics to avoid (e.g., legal, medical, HR-sensitive content)
• Actions to avoid (e.g., final approvals, publishing, deleting) without human review
• Data boundaries and privacy constraints

Example:

• "Do not provide legal or compliance advice."
• "Never send emails automatically; always create drafts only."
• "Do not export or summarize data that contains personal identifiers beyond what is visible in the current context."

7. Example Prompts and Interactions

Provide concrete examples so users understand how to use the agent:

• Starter prompts
• Good vs. bad input examples
• Typical workflows

Example prompts:

• "Turn these raw notes into a discovery summary with next steps and risks."
• "Draft a follow-up email based on the last opportunity summary."

Designing Effective Agent Identities

1. Start from Real Problems

When proposing a new agent identity, answer:

1. What recurring task is painful or slow today?
2. Who owns that task? (role/team)
3. What does a good outcome look like? (artifact, decision, action)

Agents that succeed usually:

• Replace or accelerate an existing manual workflow
• Fit naturally into current tools and processes
• Have a clear “before/after” improvement (time saved, quality, consistency)

2. Keep Scope Narrow at First

• Focus on one primary use case rather than many loosely related ones.
• Document out-of-scope tasks so users know when to switch agents.
• If an agent keeps being used for a different purpose, consider splitting into two agents.

3. Optimize for Discoverability

Ensure users can easily find the right agent:

• Use descriptive names that match user language, not internal jargon.
• Write a one-line description that appears in lists and search.
• Tag agents by team, function, or workflow.

Example listing:

• Name: Customer Support Triage Assistant
• Description: Categorizes new tickets, suggests priority, and drafts initial responses for support agents.

4. Make Behavior Predictable

• Document response formats (e.g., headings, bullet structure).
• Specify whether the agent should:
  • Ask for confirmation before major actions
  • Automatically suggest next steps
  • Provide citations or links when referencing internal documents

Example behavior rule:

Always answer with: Summary, Recommended Actions, and Risks or Open Questions.

5. Iterate with Real Usage

After launching an agent identity:

1. Collect examples of successful and failed interactions.
2. Refine behavior and constraints based on patterns.
3. Update documentation and example prompts as the agent evolves.

Example Agent Identity: Release Notes Drafter

Below is a sample agent definition to illustrate how to apply these guidelines.

Summary

• Name: Release Notes Drafter
• Tagline: Summarizes changes from engineering tickets and commits into user-facing release notes.

Target Use Case

• Primary: Convert a list of merged pull requests or completed tickets into a structured release notes draft.
• Secondary: Suggest segment-specific notes (admin vs. end-user) from the same source list.
• Out-of-scope: Roadmap planning, effort estimation, or bug triage.

Target Audience

• Product managers and technical writers preparing release communication.

Behavioral Guidelines

• Tone: Clear, neutral, and user-facing; avoid internal code names.
• Style: Group changes by feature area; highlight impactful changes first.
• Interaction: If ticket descriptions are vague, ask for clarifications before finalizing.

Capabilities and Tools

• Data: Read-only access to issue tracker and merged pull requests for a given release.
• Operations: Generate structured markdown with headings like New, Improved, Fixed.

Constraints

• Do not mention internal ticket IDs or repository names in user-facing text.
• Do not speculate about future changes.

Example Prompts

• "Generate release notes for version 2.4.0 from these JIRA tickets and Git commits."
• "Draft separate release notes for admins and end users based on this change list."

Launching and Communicating Agent Identities

To successfully launch agent identity as a capability and individual agents within it:

1. Document each agent using the template above.
2. Create a catalog or directory where users can browse available agents by use case.
3. Provide onboarding guidance:
   • Short descriptions for each agent
   • Starter prompts
   • Expected inputs/outputs
4. Set feedback channels so users can report issues, request new behaviors, or suggest new agents.

Suggestions for Expanding Agent Identity Coverage

Consider designing agent identities for:

• Go-to-market teams
  • Sales discovery summarizer
  • Proposal/quote explainer
  • Customer success QBR builder
• Product & engineering
  • Requirements clarifier
  • Test case generator
  • Incident timeline summarizer
• Operations & support
  • Ticket triage assistant
  • Macro and knowledge article suggester
• People & HR
  • Policy Q&A explainer
  • Interview feedback summarizer

For each new idea, run it through the template and ensure:

• The target use case is specific and high-value.
• The audience is clearly defined.
• Behavior, capabilities, and constraints are explicit.

Next Steps

• Use this article as a baseline to write or refine your agent identity draft.
• Create a standardized form or document based on the template.
• Pilot a small set of high-impact agent identities before scaling to more use cases.