---
title: "Build and scope agents"
slug: "build-and-scope-agents"
updated: 2025-12-16T10:18:08Z
published: 2025-12-16T10:18:08Z
canonical: "support.tulip.co/build-and-scope-agents"
---

> ## Documentation Index
> Fetch the complete documentation index at: https://support.tulip.co/llms.txt
> Use this file to discover all available pages before exploring further.

# Build and scope agents

## Building and Scoping AI Agents in Tulip

## Overview

Building and deploying AI agents in Tulip streamlines operations by automating workflows, interpreting data, and providing actionable insights. The foundation of a successful agent is a clear definition of its purpose, boundaries, and intended business value. This article offers practical steps and best practices for planning, designing, and configuring agents reliably for your environment.

---

## Why Scoping Matters

A well-scoped agent:

- **Delivers precise outcomes**, minimizing ambiguity.
- **Reduces development time** and rework by narrowing focus.
- **Simplifies testing and evaluation** with clear success criteria.
- **Builds user trust** through reliable, predictable results.

---

## Key Steps in Building & Scoping an AI Agent

### 1. Define the Agent’s Objective

Start by answering:

- What specific problem or task will this agent address?
- Who are the intended users (e.g., operators, supervisors, engineers)?
- What value or outcome should it provide?

          Tip

          

Write a one-sentence description, e.g.: “This agent generates a daily summary of shift activities for line supervisors.”

---

### 2. Set Boundaries and Constraints

Clearly describe what the agent **should** and **should not** do:

- **Included:** The types of data, actions, or queries the agent can handle.
- **Excluded:** Anything outside of its intended scope.

          Example

          

Include: Queries about work order status, inventory lookups. Exclude: Modifying user permissions, approving batch releases.

---

### 3. Outline Data Requirements

Document what input data the agent needs and what outputs it will produce:

- **Inputs:** Data tables, user prompts, context, integrations.
- **Outputs:** Reports, responses, suggested actions.
- **Access:** System privileges or data sources needed.

---

### 4. Design the Agent’s Prompt and Instructions

Draft a specific, clear prompt for the agent. Outline:

- Its primary role or goal.
- Tasks to perform.
- Behavioral guidelines (tone, format, escalation rules).
- How to handle edge cases or missing data.

---

### 5. Select and Configure Tools

List which Tulip tools, APIs, or integrations the agent needs. Set up access and permissions as required.

---

### 6. Define Test Cases and Evals Before Deployment

Develop clear test cases (“evals”) that represent real user scenarios. For each, specify:

- Input or prompt.
- Expected output.
- Criteria for success.

          
          

See the [Evaluations](https://support.tulip.co/docs/agent-evaluations) article for more.

---

### 7. Review and Iterate

Share your agent’s scope and configuration for stakeholder feedback. Test in a sandbox, collect feedback, and iterate before full deployment.

---

## Quick Reference Checklist

- Objective and user group clearly defined.
- Tasks in-scope and out-of-scope are listed.
- Input/output requirements are documented.
- Agent prompt and instructions are clear.
- Tools and permissions configured.
- Test cases (evals) are written.
- Post-launch review and feedback plan in place.

---

## Further Reading

- [AI Agent Overview](https://support.tulip.co/docs/ai-agents) - Get familiar with AI Agents in Tulip
- [Agent Evaluations](https://support.tulip.co/docs/agent-evaluations) - Learn how to create effective evaluations
- [AI Agent Library](https://library.tulip.co/overview/tulip-ai-agent-library) - Explore the full AI Agent library in Tulip