Agent Instructions

At the core of every Doti Agent lies a set of instructions - not just prompts, but operational blueprints that define how the agent thinks, searches, and responds.

Agent instructions transform a generic chat interaction into a goal-oriented AI process tailored to your business logic.

---

What Are Agent Instructions?

Agent Instructions are the configurable layer that:

1. Define the reasoning strategy Tell the agent how to search, what tools to use, and in what order.

2. Set tone and structure Guide the style, tone, and formatting of the final answer - from bullet points to executive briefings or structured tables.

3. Enable multi-step synthesis Chain together different retrieval and reasoning steps to go from scattered input → structured, actionable output.

AI and LLMs don’t inherently know the level of detail or granularity you expect in a response. By giving clear, structured instructions, you can significantly improve the quality and precision of the output.

Instead of a vague request like:

“Find opportunities”

Try something more explicit:

“Find all open opportunities with an amount greater than $30k. For each, include the description, owner, and expected timeframe.”

Clear inputs lead to powerful, tailored outputs.

---

What Instructions Can Control

Area	Example
Search Strategy	“Start by looking in Salesforce. Then check Confluence and Slack for support context.”
Process Type	“Perform each step sequentially and synthesize insights after all data is gathered.”
Output Formatting	“Summarize findings in three sections: Problem, Root Cause, Resolution.”
Tone	“Use a formal tone and avoid speculative language.”
Tools	Specify if the agent should use web search, CRM, or documents only

---

Example: Complex Multi-Step Workflow

Scenario: A ticket arrives and we want to fully triage it.

Instruction Blueprint

For every new support ticket:

1. Search for similar past tickets based on the current ticket content.

2. Search internal knowledge bases (Confluence, Drive, Notion) for possible solutions.

3. Summarize the ticket in the following format:

   • Issue:

   • Description:

   • Potential Resolution:

4. Categorize the ticket using this predefined list of categories:

   • Login Issue

   • Integration Bug

   • Data Mismatch

   • Feature Request

   • Other

5. Generate a final summary combining all findings and proposed resolution steps, and tag the responsible team.

---

Combining Techniques

One of the most powerful aspects of instructions is that they allow compound reasoning and tool chaining. In a single instruction set, you can tell the agent to:

• Search across tools (Slack, Salesforce, Zendesk)

• Apply formatting logic (e.g. tables, markdown, summaries)

• Trigger follow-up actions (e.g. summarize, categorize or draft)

• Maintain grounding in scoped access

🧠 Instructions are not static prompts - they are dynamic planning guides that shape how the agent works in your org.

---

More Real-World Instruction Use Cases

Customer Preparation Agent

Prepare a structured, actionable meeting brief for a specific customer using internal data across systems.

Workflow Steps:

1. Identify the customer by name and ensure all retrieved data is scoped to this account only.

2. Search Salesforce for account details, opportunities, and key contacts.

3. Search Gong for relevant calls and extract recent conversation highlights.

4. Search Slack and Drive for recent internal discussions or documents mentioning this customer.

5. Correlate all retrieved data to build a customer snapshot.

Output Format:

• 1. Customer Overview

  • Company Name

  • Industry

  • Key Contacts (name + title)

  • ARR (from Salesforce)

  • Notable News (from Slack or web)

• 2. Opportunities & Critical Cases

  • Open Opportunities: Name, Amount, Stage

  • Open Support Cases: Subject, Priority, Created Date

• 3. Relationship Summary

  • Past 3 Interactions (Gong or Slack): Who, When, What

  • Key Trends / Sentiment

• 4. Meeting Objectives

  • Customer's Goals

  • Our Goals

  • Likely Challenges/Objections

• 5. Gong Call Insights (if available)

  • Last 3 call summaries with key points, speakers, tone, action items

• 6. Key Internal Docs

  • Document Title, Description, Link (from Google Drive)

• 7. Slack Recap

  • Channels Discussing the Customer

  • Sentiment Summary

  • Action Items Raised

Tone & Structure: Professional, to-the-point, formatted for executive review or CSM use. Use bullets and section headers.

---

Weekly Digest Agent

Create a weekly summary of what happened across teams, focused on key updates, blockers, and wins.

Workflow Steps:

1. Search Slack for important discussions in channels like #gtm, #product, #sales, etc.

2. Search Confluence and Notion for newly published or updated docs tagged with “weekly” or “update”.

3. Check Gong call summaries for any major insights from customer conversations.

4. Filter out low-signal chatter or repetitive updates.

5. Cluster findings by department/team.

Output Format:

• 1. Executive Summary

  • 3–5 top-line highlights across teams

• 2. Product

  • Key releases

  • Bugs resolved

  • Upcoming features under discussion

• 3. Sales

  • New wins

  • Major pipeline movements

  • Objection trends or themes

• 4. Customer Success

  • Escalations closed

  • Onboarding challenges

  • Customer feedback trends

• 5. Marketing

  • Campaign launches

  • Performance insights

  • Collateral created or updated

• 6. Risks & Blockers

  • Any cross-team dependencies or red flags

Tone & Structure: Concise, bullet-point format. Use markdown styling (headers, bold) for easy readability in Slack or email.

Marketing Agent

Assist in producing marketing content using internal data and recent updates.

1. Search for the latest product updates from Notion and Slack tagged #product-updates

2. Identify recent customer wins or testimonials from Salesforce and Gong summaries

3. Extract positioning language or value props from GTM docs

4. Format the output into a brief with these sections:

   • Headline Idea

   • Customer Win Highlights

   • Product Messaging

   • Suggested CTA or Channel

5. Tone should be clear, concise, and suitable for social, blog, or campaign use

---

Summary

What Instructions Do	Why It Matters
Plan the search strategy	Gives structure to the agent’s process
Guide step-by-step reasoning	Ensures multi-source, multi-hop workflows
Control tone & output format	Tailors response to the right audience
Enable complex tasks	From triage to synthesis to final summary
Support modular reuse	Use similar instructions across agents or APIs

---

✨ Think of Agent Instructions as operational playbooks. With them, Doti doesn’t just answer - it understands, executes, and communicates the way your team needs.