How to Build Your First Agentforce Agent - A Complete Beginner's Guide With Prerequisites & Step-by-Step Setup

Salesforce just changed what it means to be an Admin.

For the first time, a Salesforce Admin can build an AI agent that reasons, takes action, and works alongside employees — without writing a single line of code.

The people learning Agentforce today may become the AI Architects of tomorrow.

In this edition, I'm walking you through exactly what Agentforce is, what you need before you start, and the complete step-by-step process to build your first Agentforce Employee Agent — the internal AI assistant that helps your own team get work done faster inside Salesforce.

Here's how to build your first agent in under an hour.

By the end of this, you'll have a working agent in your own org. Let's go.

What Is Agentforce?

Agentforce is Salesforce's platform for building autonomous AI agents — software that can reason through a problem, decide what to do, take action, and respond, without a human typing a prompt for every single step.

This is a real shift from older tools:

A traditional chatbot follows a fixed script. It can only say what it was pre-programmed to say. Einstein Copilot (the predecessor) waits for a human to ask it something, then responds. An Agentforce agent can be triggered by an event — a new case, an incoming email, a record update — and independently work through a multi-step task to resolution.

Think of it like this: a chatbot is a vending machine. It only gives you what's pre-programmed. An Agentforce agent is more like a new team member who reads the situation, reasons about what's needed, and takes the right action.

Under the hood, every agent runs on the Atlas Reasoning Engine — the AI "brain" that interprets what a user is asking, decides which Topic and Action applies, and executes it using your real Salesforce data.

Agentforce Studio is your home base. Agent Builder is where you configure the agent. Plan Tracer shows you its reasoning in real time. Conversation Preview is where you test it before going live.

🧩 THE TWO TYPES OF AGENTFORCE AGENTS

Before you build anything, understand the two categories:

→ EMPLOYEE AGENTS (Internal Agents)

Internal-facing. Helps your own employees — sales reps, service agents, ops teams — get information and complete tasks faster inside Salesforce. This is the direct successor to Einstein Copilot.

Examples

Sales Assistant Agent Service Agent Assistant HR Support Agent Operations Agent CRM Data Assistant

Common Use Cases

✅ Summarize Opportunities

✅ Create or update records

✅ Retrieve customer information

✅ Generate emails

✅ Answer internal process questions

✅ Analyze pipeline data

Where They're Used

Salesforce Lightning Experience Salesforce Mobile App Internal employee workflows

Think of Employee Agents as the evolution of Einstein Copilot.

→ CUSTOMER-FACING AGENTS or SERVICE AGENTS (External Agents)

These agents interact directly with customers. Handles support queries, case deflection, appointment booking, and FAQ resolution for your customers — deployed to websites, WhatsApp, Messaging, and Experience Cloud portals. Today's focus: the Employee Agent — the best starting point for anyone new to Agentforce, because it works entirely inside your own org with your own team as the test audience.

Examples Customer Support Agent FAQ Agent Appointment Booking Agent Order Tracking Agent Self-Service Agent

Common Use Cases ✅ Answer support questions

✅ Create cases

✅ Reset passwords

✅ Book appointments

✅ Check order status

✅ Deflect support tickets

Where They're Used Company websites Experience Cloud portals Messaging channels WhatsApp Mobile applications

🧱 THE THREE BUILDING BLOCKS OF EVERY AGENT

Every Agentforce agent — no matter how simple or complex — is made of three things:

→ TOPICS : The subject areas your agent is allowed to talk about. A Topic has a name, a description, instructions, and a scope. Topics keep your agent focused and stop it from wandering off-task.

→ ACTIONS : What the agent actually DOES. Actions can be Standard (pre-built by Salesforce, like Query Records or Summarize Record) or Custom (built by you using Flow, Apex, or Prompt Templates).

→ INSTRUCTIONS : Natural-language guidance you write that tells the agent how to behave within a Topic — what tone to use, what data to reference, what NOT to do. Build order matters: Agent → Topic → Actions inside that Topic → Instructions for that Topic. Start with one Topic. Expand from there.

✅ PREREQUISITES — WHAT YOU NEED BEFORE YOU START

Check every box below before you open Agent Builder.

THE RIGHT SALESFORCE EDITION & LICENSE

You need access to an Enterprise, Performance, or Unlimited Edition org with Agentforce enabled.

🆓 Don't have access? Spin up a free Agentforce Developer Edition org instead — it comes with Agentforce and Data Cloud pre-enabled, completely free. → developer.salesforce.com/signup

ENABLE THE REQUIRED SETUP FEATURES

Before building an agent, make sure these are turned on on in Setup:

→ Einstein Setup — turned ON

→ Einstein Bots / Agents — enabled

→ Embedded Service Deployment — enabled (if you'll deploy to a Lightning page)

→ Messaging — enabled (if relevant to your use case)

How to check:

Setup → Quick Find → "Einstein Setup" → toggle ON

Setup → Quick Find → "Agents" → turn on the Agentforce toggle

AGENTFORCE (DEFAULT) AGENT ENABLED

If you previously used Einstein Copilot, you're already partway there — existing Copilot configurations can migrate into Agentforce Topics and Actions.

Setup → Quick Find → "Agents" → find "Agentforce (Default)" → turn the toggle ON.

If you hit an error opening Agent Builder, it's usually because no default/employee agent has been activated yet — activate one first, then proceed.

CLEAN, ACCESSIBLE DATA

Your agent can only answer using data it has access to. Before building:

→ Identify which objects your agent needs to query (e.g., Cases, Opportunities, Knowledge Articles)

→ Make sure field-level security and sharing rules allow the agent's running user to see that data

→ If your agent needs unstructured data (PDFs, docs), prepare a Data Library to upload them into

THE RIGHT PERMISSIONS FOR THE RUNNING USER

Every agent runs as a specific user in the background — its "Agent User." This user needs:

→ A permission set granting access to the objects and actions the agent will use

→ Field-level security aligned to what the agent should and shouldn't see

This is a critical, often-missed step — an agent is only as capable as the user it runs as.

A CLEAR JOB DEFINITION (DO THIS ON PAPER FIRST)

Before touching Agent Builder, answer these three questions:

→ Who is this agent for? (which team or role)

→ What specific job will it do? (one focused job — not "everything")

→ What data and actions does it need to do that job?

Remember: you don't want your agent to do EVERYTHING — just like you wouldn't want one employee to do everything. Pick one focused job for your first agent.

🛠️ STEP-BY-STEP: BUILD YOUR FIRST EMPLOYEE AGENT

We'll build a simple, real example: an Employee Agent that helps Sales reps quickly summarize an Opportunity and check for related open Cases on the same Account — a genuinely useful first agent.

STEP 1 → Open Agentforce Studio

→ Click the App Launcher (the grid icon, top left)

→ Search for "Agentforce Studio" and open it

→ You'll land on the Build pane — this is your home for all agents, analytics, data libraries, and testing

Think of Agentforce Studio as Setup for AI: Agent Builder is to Agentforce what Flow Builder is to Salesforce Flow.

STEP 2 → Confirm the Agentforce (Default) Agent is active

→ In Agentforce Studio, find the Agents list

→ Confirm "Agentforce (Default)" shows as Active

→ If not active yet, toggle it ON now (this is your Employee Agent base)

STEP 3 → Open the agent in Agent Builder

→ Click on "Agentforce (Default)" in the agents list → Click "Open in Builder" (top right) → You'll now see three panels:

• LEFT: Topics, Actions, and Instructions you can customise

• CENTER: Plan Tracer — shows the agent's reasoning step by step

• RIGHT: Conversation Preview — where you chat with your agent to test it live

STEP 4 → Remove the default placeholder topic (optional but recommended)→

Find "MigrationDefaultTopic" (or similar default topic) in the Topics list → Click the dropdown arrow next to it → select "Remove from Agent"

This keeps your first build clean and focused on the one job you defined.

STEP 5 → Create your first Topic

→ Click the dropdown next to "Manage the topics assigned to your agent" → Select "New Topic" → You have two paths here:

• Describe what you want in plain English, and Agentforce will use AI to scaffold the Topic for you

• Or build it manually from a blank Topic

For our example, type something like: "Help sales reps summarize an Opportunity and find related open Cases on the same Account"

→ Click Next — Agentforce will generate a Topic Label, Description, and Classification Description (this tells the reasoning engine WHEN to use this topic)

STEP 6 → Add Actions to your Topic

→ Inside your new Topic, go to the Actions tab → Click "New Action" (or accept the AI-suggested ones) → Choose:

• Standard Action — pre-built, e.g. "Query Records" or "Summarize Record" — fastest way to start

• Custom Action — built from a Flow, Apex class, or Prompt Template — for anything Salesforce doesn't provide out of the box

For our example, add: → Query Records (to find related open Cases on the Account) → Summarize Record (to summarize the Opportunity)

STEP 7 → Write clear Instructions for the Topic

→ Go to the Instructions field for your Topic → Write natural-language guidance, e.g.:

"When a user asks about an Opportunity, first summarize the Opportunity's key details. Then check if there are any open Cases on the same Account, and mention them. Keep responses concise and professional."

💡 Tip: All instructions for a topic load into one instruction field, in order. The order matters — write them like a checklist your agent should follow.

STEP 8 → Connect your Data (if needed)

If your agent needs unstructured data (PDFs, knowledge docs):

→ Go to Data Libraries in the left panel → Click to create a new Data Library → Set Data Type to "Files" → Upload your documents → Click Done — your agent can now reference this content when reasoning

STEP 9 → Test your agent in the Conversation Preview

→ In the right-hand Conversation Preview panel, type a real question: "Summarize Opportunity [Name] and tell me if there are any related open cases"

→ Watch the Plan Tracer in the center panel — it shows exactly which Topic was selected, which Actions were called, and how the response was built

This transparency is one of Agentforce's best features — you can see the agent's reasoning, not just its answer.

STEP 10 → Assign the right Permission Set

→ From Setup, search "Permission Sets" → Find the permission set associated with your agent (e.g., a Service/Employee Agent permission set) → Click "Manage Assignments" → Assign it to the agent's running user

Without this step, your agent will be reasoning correctly but failing to access the data it needs — one of the most common first-time issues.

STEP 11 → Activate your agent

→ Back in Agent Builder, click "Activate" (top right corner) → Your agent is now live and usable inside your org

STEP 12 → Try it for real

→ Open any record where your agent applies (e.g., an Opportunity) → Click the Einstein icon (top right of the page) to open the Agent sidebar → Ask your agent the question you tested earlier — now in a real working context

You just built and deployed your first AI agent. No code. No data science team. ⚠️

COMMON FIRST-TIME MISTAKES

❌ Trying to make one agent do everything → Start with ONE focused job. Expand with more Topics later.

❌ Forgetting to assign the permission set to the running user → Your agent will "reason" correctly but fail to retrieve data. Always assign permissions before testing.

❌ Vague Topic descriptions → The Classification Description tells the reasoning engine WHEN to use a Topic. Vague descriptions cause the agent to pick the wrong Topic.

❌ Skipping the Conversation Preview / Plan Tracer → This is your debugging tool. Always test here before activating — it shows you exactly why the agent responded the way it did.

❌ Not removing the default placeholder topic → Leftover default topics can confuse the reasoning engine when multiple topics overlap.

📚 FREE RESOURCES TO GO DEEPER

🆓 Trailhead: "Introduction to Agent Builder" trailhead.salesforce.com → search "Agent Builder"

🆓 Trailhead: "Agentforce" official Trailmix trailhead.salesforce.com → search "Agentforce"

🆓 Free Agentforce Developer Org (Agentforce + Data Cloud built in) developer.salesforce.com/signup

🆓 Salesforce Developer Workshop — Build a Service Agent developer.salesforce.com/workshops/agentforce-workshop

🆓 Agentblazer Status Program trailhead.salesforce.com/agentblazer

💬 FINAL THOUGHT

Six months ago, building an AI agent felt like something reserved for engineering teams.

Today, it's a Tuesday afternoon project for any Salesforce Admin willing to follow these 12 steps.

The professionals who build their first agent now — even a simple one — will be far ahead of the curve when their organisation asks "can we automate this with AI?"

Start small. One Topic. One job. Build from there.

Question for You Have you built your first Agentforce agent yet?

Yes — already live in my org Building one this week after reading this Still exploring — haven't started yet

Drop your stage in the comments. If you get stuck on any step, ask — I'll help troubleshoot.

Enjoyed this guide? Salesforce Pulse publishes practical Salesforce, Agentforce, AI, Data Cloud, and Architecture tutorials every week.

Subscribe so you don't miss future editions.

Written by Dollesh Rathod · Salesforcebydollesh

Part of the Salesforce Pulse Newsletter series

#Salesforce #Agentforce #EmployeeAgent #AgentBuilder #SalesforceAdmin #SalesforceDeveloper #AIAgents #AgentforceTutorial #AtlasReasoningEngine #Trailblazer #SalesforcePulse #Agentblazer #CRM

Follow on LinkedIn :