How to Interact with an AI Agent

Getting great results from your AI agents is all about how you communicate. This guide shows you how to give effective instructions and use the chat interface effectively.

Understanding the Chat Interface

Before diving into best practices, let's walk through the chat interface so you know where everything is.

For a detailed guide on all interface elements, see Navigating FabriXWork.

Sidebar (Left Panel)

The sidebar shows all your agents:

• Pinned agents appear at the top (agents you've starred)
• All other agents are listed below
• Click any agent to switch to their chat
• Star an agent to pin them to the top for quick access

Header (Top Section)

The header displays the current agent's information:

• Agent name and role
• Click the header to view full agent details
• Star icon — Click to pin/unpin the agent in your sidebar
• Agent Profile — Access agent-specific configurations

Chat Area (Center)

The main conversation space:

• Message history — All your previous conversations with this agent
• Chat input box — Type your instructions at the bottom
• Send button — Submit your instruction
• File attachments — Add files directly to your message (if supported)

Folder Structure (Right Panel)

The right panel shows your connected workspace:

• Folder tree — Navigate your connected folder structure
• File list — See all files the agent can access
• File preview — Click files to preview content
• Agents can read and (in Build Mode) modify these files

Configure Your Agent Before Asking

Before sending your first instruction, you can adjust several settings to control how the agent works.

1. Plan Mode vs Build Mode

One of the most important settings is choosing between Plan Mode and Build Mode.

Best Practice: Plan First, Build Later

A recommended workflow is to start with Plan Mode, then switch to Build Mode:

Step 1: Plan Mode

"Review the files in my Fundraising Campaign folder and suggest a structure 
for organizing our campaign materials. What files should we create?"

Step 2: Review the plan

• Agent provides recommendations
• You review and refine the approach
• No files are changed yet

Step 3: Switch to Build Mode

"Great plan! Now create the folder structure and draft the initial documents 
we discussed."

This approach helps you:

• ✅ Understand what the agent will do before it makes changes
• ✅ Refine the approach without risking unwanted file modifications
• ✅ Get multiple plan options before committing to one

2. Choose AI Model

Different AI models have different strengths:

• Faster models — Quick responses for simple tasks
• Advanced models — Better reasoning for complex analysis
• Specialized models — Optimized for specific tasks (coding, writing, etc.)

How to select:

1. Click the model selector in the chat header
2. Choose a model based on your task complexity
3. Consider speed vs quality trade-offs

3. Configure Thinking Method

Control how deeply the agent thinks before responding:

Recommendation: Start with Default, then increase depth if the agent's responses seem superficial for your task.

4. Session Configuration (Before You Start)

When you first connect an agent to a folder, you'll see the "Configure Your Session" modal:

This modal helps you set clear expectations before starting your work together.

"What would you like help with?" Field

This optional field helps you set context for the agent before starting:

Purpose:

• Gives the agent a head start on understanding your goals
• Helps the agent prepare relevant context from your files
• Sets the tone for the conversation

Good Examples:

I need help creating a product requirements document for a new 
feature that tracks employee attendance.

Please analyze our Q1 sales data and identify trends we should 
present to the board.

Help me draft a fundraising campaign proposal for our NGO.

Bad Examples:

Help me

Too vague—the agent doesn't know what you need.

Do stuff

No context provided—the agent can't prepare appropriately.

Tips:

• ✅ Be specific about your task or goal
• ✅ Mention the type of output you need
• ✅ Reference key files or data if relevant
• ❌ Don't worry about perfect wording—the agent will ask clarifying questions

Session Summary

The modal shows a summary of your configuration:

Mode Indicators:

• 📋 Plan Mode Only — Agent can analyze and discuss, but cannot modify files
• 🔨 Build Mode — Agent can read AND modify files in your connected folder

What Happens After You Click "Start Chat"?

1. Agent loads your context — The agent scans your connected folder
2. Session initializes — Your conversation history starts fresh
3. Agent is ready — The agent can now access files and respond to your instructions

Note: If you provided a task description, the agent may:

• Acknowledge your goal
• Ask clarifying questions
• Start working on your request immediately

When to Use Session Configuration vs Skip

The Basics of Agent Interaction

Agents in FabriXWork are like skilled teammates—they have expertise, but they need clear direction from you.

The Interaction Flow

1. You give an instruction
         ↓
2. Agent processes your request
         ↓
3. Agent asks clarifying questions (if needed)
         ↓
4. Agent executes the task
         ↓
5. Agent delivers output
         ↓
6. You review and iterate (if needed)

Good vs Bad Instructions

❌ Bad Instruction (Too Vague)

"Make a presentation"

Problems:

• No topic specified
• No audience defined
• No format or structure requested
• Agent doesn't know what to do

✅ Good Instruction (Clear & Specific)

"Create a 10-slide presentation with frontend-slides about our Q1 marketing results for the executive team.
Include slides on: campaign performance, budget vs actuals, key wins, and lessons learned.
Use a professional tone and include data visualizations where possible."

Why it works:

• ✅ Clear topic (Q1 marketing results)
• ✅ Defined audience (executive team)
• ✅ Specific structure (10 slides, specific topics)
• ✅ Tone guidance (professional)
• ✅ Format preference (data visualizations)

Key Principles for Effective Instructions

1. Provide Context

Without context:

"Write a report"

With context:

"We just completed the customer satisfaction survey for Q1. 
Write a summary report highlighting the top 3 improvements and 2 areas needing attention."

What context helps:

• Background information
• Why this task matters
• Who will use the output
• Any constraints or requirements

2. Be Specific About Output Format

Vague:

"Give me the data"

Specific:

"Provide the data as a table with columns: Name, Department, Start Date, and Manager.
Sort by Department, then by Name."

Format options you can request:

• Table
• Bullet points
• Numbered list
• JSON format
• Paragraph summary
• Email draft
• Presentation outline

3. Ask for Structured Output

Unstructured:

"Tell me about the project status"

Structured:

"Provide a project status update with these sections:
1. Completed This Week
2. In Progress
3. Blockers
4. Next Week's Priorities
5. Risks"

4. Iterate for Better Results

Your first instruction might not be perfect—that's okay! Agents are designed for iteration.

Example iteration flow:

First attempt:

"Create a list of competitors in the fintech space"

Agent provides a basic list

Refinement:

"Good start! Now organize them by category (payments, lending, wealth management).
For each competitor, add: founding year, funding raised, and key differentiator."

Second refinement:

"Perfect! Now export this as a table and save it to my Research folder as 'Fintech_Competitors.xlsx'"

Practical Examples

Example 1: Document Analysis

Task: Review a contract and summarize key points

Instruction:

I'm uploading a vendor contract. Please:
1. Identify the contract type and parties involved
2. List all payment terms and amounts
3. Highlight any auto-renewal clauses
4. Flag any unusual terms or potential risks
5. Summarize in bullet points

Save your analysis to the Contracts folder.

Example 2: Data Extraction

Task: Extract information from multiple files

Instruction:

Review all the interview transcripts in the Research folder and:
1. Extract all mentions of "pricing concerns"
2. Group them by customer segment (Enterprise, SMB, Startup)
3. Create a summary table with: Segment, Concern, Frequency, Quote
4. Identify the top 3 pricing concerns overall

Output as a formatted table.

Example 3: Content Creation

Task: Write an email

Instruction:

Draft a follow-up email to attendees of yesterday's webinar.

Include:
- Thank them for attending
- Link to the recording (use placeholder: [RECORDING_LINK])
- Link to the slides (use placeholder: [SLIDES_LINK])
- Mention the next webinar date: March 20, 2026
- Call-to-action: Schedule a demo

Tone: Friendly and professional
Length: Under 200 words

Providing Context Files

Agents can work with files in your connected folders.

How to Reference Files

Option 1: Direct reference

"Use the Q1_Report.docx file in my Reports folder to create the summary"

Option 2: Folder reference

"Review all files in the Customer_Feedback folder and identify common themes"

Option 3: Let the agent explore

"Look at the files in my current workspace and tell me what data is available"

Asking Clarifying Questions

If you're not sure how to phrase something, ask the agent!

Examples:

"What information do you need from me to create this report?"

"Is there a better way to structure this request?"

"What format would work best for this output?"

Common Mistakes to Avoid

Quick Reference: Instruction Templates

For Document Tasks

"Review [FILE/FOLDER] and [ACTION]. Output as [FORMAT] with [SECTIONS/STRUCTURE]."

For Data Tasks

"Extract [DATA TYPE] from [SOURCE]. Organize by [CRITERIA]. Output as [FORMAT]."

For Content Creation

"Create [CONTENT TYPE] about [TOPIC] for [AUDIENCE]. Include [ELEMENTS]. Tone: [TONE]."

For Analysis Tasks

"Analyze [DATA/SUBJECT] and identify [WHAT TO FIND]. Present findings as [FORMAT]."

Session Configuration Quick Tips

When to Configure Session