Build an Agent That Thinks Like You
A well-configured agent doesn't just complete tasks — it understands your context, follows your guidelines, and produces output you'd be proud to share with a client.
What You'll Learn
- Create a well-scoped agent with a clear role and constraints
- Write effective system prompts that produce consistent, high-quality output
- Choose the right model for your agent's task type
- Test and iterate on agent configuration using the Chat interface
- Reuse agents across multiple workflows
Agents are the building blocks of every FlowManner workflow. Each agent combines a model, a system prompt, and a set of tools into a reusable unit that you can chain, parallelize, or delegate to.
Prerequisites
- FlowManner account with at least one API key configured (see Setting Up BYOK)
- Basic familiarity with what a system prompt is
- A specific task you want the agent to handle
What Makes a Good Agent
The best agents have three things: a clear role (what they do), specific constraints (what they don't do), and access to the right tools. Think of it like onboarding a new team member — the more context you give upfront, the less supervision they need later.
# Example System Prompt for a Client Report Writer
You are a business report writer specializing in competitive analysis.
## Role
- Analyze competitive intelligence data and produce executive-ready reports
- Focus on actionable insights, not raw data dumps
## Output Format
- Use markdown with these sections: Executive Summary, Key Findings,
Trend Analysis, Recommendations
- Keep each section under 200 words
- Use bullet points for findings, paragraphs for analysis
## Constraints
- Never speculate without labeling it as "Hypothesis"
- Always cite the data source for each finding
- Avoid jargon — write for a non-technical CEO audience
- If data is insufficient, say so explicitly rather than filling gaps
Choosing a Model
Match the model to the task. GPT-4o excels at analysis and structured output. Claude 4 Sonnet handles long documents and nuanced writing. DeepSeek offers strong reasoning at lower cost. Start with the default model and experiment once you're comfortable.
System Prompt Best Practices
Write system prompts like brief emails to a smart colleague: state the role, the audience, the output format, and any hard constraints. Avoid vague instructions like "be helpful" — instead, specify exactly what "helpful" means in your context.
Tool Access and Permissions
Each agent can be granted access to specific tools: web search, file operations, code execution, or API calls. Grant only the tools the agent needs — a report writer doesn't need code execution. You can always add tools later if the agent's scope expands.
Iterating on Prompts
Your first system prompt won't be perfect. Test with 3-5 real tasks, examine the output critically, and refine. Common fixes: adding examples of desired output, tightening constraints, specifying edge case handling. Most agents reach production quality after 2-3 iterations.
Start narrow
Give your agent a single, well-defined responsibility. "Write email subject lines" beats "Handle all email tasks" — narrow scope means higher quality.
Include output format in the prompt
Always specify how you want the output structured: bullet list, markdown table, JSON, numbered steps. Ambiguous format means inconsistent results.
Pre-built agents available
New accounts come with ready-to-use agents for common tasks. Customize them instead of building from scratch — it's faster and you'll learn the patterns.
Step-by-Step
Open the Agents Page
Click Agents in the sidebar to see your agent library. New accounts start with a set of pre-built agents you can use as-is or customize.
Click Create Agent
Hit the Create Agent button in the top right. You'll see a form with Name, Description, Model, and System Prompt fields.
Name and Describe Your Agent
Give it a descriptive name like 'Client Report Writer' or 'Code Reviewer'. The description helps you remember what this agent does when you have dozens of them.
Write the System Prompt
Define the agent's role, the format of its output, and any constraints. For example: 'You are a technical writer. Produce markdown documentation with code examples. Keep explanations under 200 words per section. Never use jargon without defining it first.'
Test in the Chat
Open the Chat page, select your new agent from the dropdown, and give it a real task. Iterate on the system prompt until the output matches your expectations.
Related guides
Ready to start building?
Practical tutorials for getting the most out of FlowManner — from your first workflow to advanced multi-agent orchestration.
Start building — it's free