How to Set Up an Agentforce Agent Without Code in Salesforce 2026 (Step-by-Step)

If you've been wondering how to set up an Agentforce agent in Salesforce without writing a line of code, this guide walks through the full process. Agentforce Builder is Salesforce's point-and-click interface for creating AI agents—no Apex, no developer, no platform-engineering ticket required. By the end of this walkthrough, you'll have a working service agent configured, tested, and ready to deploy.

This guide is for admins who know their way around Setup and Lightning Experience but haven't touched Agentforce yet.

How Agentforce Builder Works (and Why Admins Can Set This Up)

Agentforce Builder is a declarative configuration tool inside Salesforce Setup. You describe what you want the agent to do—using plain English instructions, not code—and the platform's reasoning engine figures out how to route conversations and execute tasks.

The no-code claim is real, with one important caveat: if you want the agent to do something beyond answering questions—like creating a case record or looking up an order—you'll need at least a simple Salesforce Flow or a standard out-of-the-box action wired up. The good news is that Flow Builder is also no-code, and Agentforce ships with a library of pre-built actions that cover common service tasks without any Flow required.

Think of Agentforce Builder as having three layers:

1. The agent itself — its name, persona, and general instructions
2. Topics — categories of intent the agent understands (e.g., "Help with billing")
3. Actions — what the agent actually does within each topic (e.g., "Look up account balance")

You configure all three layers declaratively.

Prerequisites for Setting Up an Agentforce Agent in Salesforce

Before touching the builder, make sure your org is ready. Skipping this checklist is the most common reason first-time setups stall.

Edition and Licensing

Agentforce requires Enterprise, Performance, Unlimited, or Developer Edition. Starter and Professional Edition orgs cannot use Agentforce without a specific add-on—check with your Salesforce account team if you're on one of those editions.

For customer-facing service agents, you'll need an Agentforce Service Agent license (sold separately from base CRM licenses). For employee-facing agents—tools that help your internal team—there's a separate license model. The free Agentforce Developer Edition at developer.salesforce.com includes a full Agentforce environment and is the right place to practice before touching a production org.

Pricing as of Spring '26 starts at approximately $2 per customer conversation for the per-conversation model, or flat-rate options for employee agents. These are list prices — contracted pricing varies. For exact numbers, consult your account executive before budgeting a production deployment.

Enable Einstein

In Setup, search for Einstein and make sure Einstein features are turned on for your org. This is the platform-level switch that unlocks AI capabilities across all Salesforce products, including Agentforce.

Enable Agentforce

With Einstein enabled, search for Agentforce in Setup. You should see an Agentforce section where you can enable the feature and, depending on your org, activate a default agent. If you don't see Agentforce in Setup at all, your edition or license doesn't include it yet.

Assign Permissions

Users who will build agents need specific permission sets assigned. The key ones, confirmed in org data:

• AgentforceServiceAgentBuilder — assign this to the admin who will configure agents in Agentforce Builder. This is the primary builder permission set.
• AgentPlatformBuilder — grants broader Agentforce platform builder access. Assign if you're managing the agent platform more widely.
• AgentforceServiceAgentBase — base permissions required for the service agent to function correctly at runtime. Required in addition to the builder permission set.
• AgentforceServiceAgentUser — assign to end users who will interact with the deployed agent.

In addition to these permission sets, the admin's profile or a separate permission set must include the Manage Bots user permission (or the broader Customize Application permission) to access Agentforce Builder in Setup. The Manage Bots permission is the more targeted option — it grants agent-building access without opening up all of Salesforce customization.

Tip: Assign permissions in a sandbox first and verify access before replicating to production.

Channel Setup (Required for Deployment, Not Building)

You don't need a deployment channel to build or test an agent—but you do need one before you can activate it for real users. Decide now whether you're targeting:

• Experience Cloud (a customer portal or public site)
• Messaging for In-App and Web (embedded chat on any website)
• Slack (internal employee agents)

Each channel requires its own setup steps outside Agentforce Builder. If you don't have Experience Cloud provisioned, start a sandbox experiment first.

---

Step 1: Navigate to Agentforce Builder

1. Log in to your org and go to Setup (gear icon → Setup).
2. In the Quick Find box, type Agents.
3. Under the Agents section, click New Agent or open an existing agent to edit it.

Agentforce Builder opens in a full-screen interface divided into three panels:

• Left panel (Explorer): All the agents in your org, plus their Topics, Actions, and Variables
• Center panel: The editing surface where you configure Topics and write instructions
• Right panel: The live testing console where you chat with your agent as you build

Knowing the three panels before you start saves confusion later.

---

Step 2: Create a New Agent

Click New Agent. You'll be prompted to choose a starting point.

Using a Template

Salesforce provides pre-built agent templates for common use cases—customer service, order management, HR, and others. Templates give you a working agent with pre-configured Topics and Actions that you can customize. For a first agent, start with a template. It shows you what a properly configured agent looks like and gives you something to test immediately.

Select the template closest to your use case, give the agent a name, and click Create.

Starting from Scratch

If your use case doesn't match any template, select New Agent (blank). You'll set:

• Agent Name — what the agent is called internally
• Agent Description — a brief description of what it handles (this helps Salesforce route conversations correctly in multi-agent setups)
• Agent Instructions — a set of plain-language rules the agent follows in every conversation, regardless of topic (e.g., "Always ask for the customer's account number before looking up records" or "Do not discuss pricing—direct pricing questions to a human rep")

Write your agent instructions as you would write instructions for a new employee: clear, specific, and complete. Vague instructions produce inconsistent agent behavior.

---

Step 3: Define Topics

Topics are the core routing mechanism of an Agentforce agent. When a customer sends a message, the agent's reasoning engine classifies that message against all configured Topics and activates the one that matches best.

Each Topic has:

• Topic Label — a short internal name
• Classification Description — a natural-language description of what kinds of messages belong in this topic. This is what the reasoning engine uses to decide if a message fits. Write it as you would explain the topic to a new team member: "This topic handles questions about billing, payment methods, invoice disputes, and refund requests."
• Scope — what the agent should and should not handle within this topic
• Instructions — step-by-step guidance for how the agent should behave once it's in this topic

Practical Example: A "Case Submission" Topic

Here's what a well-configured Topic looks like for a basic customer service agent that lets customers open support cases:

Topic Label: Submit a Support Case

Classification Description: The customer wants to report a problem, submit a support request, or get help with an issue that requires follow-up from the support team.

Scope: Handle new case submissions only. Do not attempt to update or close existing cases in this topic.

Instructions:

1. Ask the customer to describe their issue in a sentence or two.
2. Confirm the product or service area affected.
3. Create a new Case record with the customer's account linked.
4. Provide the customer with the case number and expected response time.
5. If the customer expresses urgency or mentions a safety issue, escalate the priority to High and notify the on-call rep.

These instructions are business logic, not code. The agent reads them and executes accordingly.

---

Step 4: Add Actions

Actions are what the agent does inside a Topic. Every Topic needs at least one Action, or the agent can only have a conversation—it can't actually update records, look things up, or trigger processes.

Actions come in three types:

1. Standard actions — out-of-the-box capabilities Salesforce provides (create a case, look up a contact, search knowledge articles). These require no configuration beyond enabling them.
2. Flow actions — any active Salesforce Flow you've built becomes available as an agent action. This is how you add custom business logic without Apex.
3. Prompt Template actions — for actions that primarily involve generating text responses using your org's data.

If you want to skip custom action-building entirely, the AgentExchange marketplace lists pre-built agents with actions already configured — a useful shortcut for common service scenarios.

To add an Action to your Topic:

1. With your Topic open in the builder, click Add Action.
2. Browse or search the action library.
3. Select an action and configure its inputs — for example, which Flow variables should be populated from the conversation.
4. Add action-level instructions if the agent needs guidance on when or how to use this specific action.

For the "Submit a Support Case" example above, you'd add the standard Create Case action and map the customer's description, account, and priority level from the conversation context.

Tip: Keep your Topics focused and your Actions specific. An agent with five well-scoped Topics and two Actions each outperforms one catch-all Topic with fifteen Actions every time. Broad Topics increase classification errors; too many Actions in one Topic confuse the reasoning engine about which one to use.

---

Step 5: Test the Agent Before Activating

The right panel of Agentforce Builder is a live testing console. Use it extensively before deploying to any real channel.

How to test effectively:

• Type messages as a real customer would, not as a system test. "My order hasn't arrived" is a better test than "Trigger Case Creation Flow."
• Deliberately send ambiguous messages to see how the agent classifies them. If it picks the wrong Topic, refine your Classification Descriptions.
• Test edge cases: What happens if the customer provides incomplete information? Does the agent ask a follow-up, or does it fail silently?
• Test the "out of scope" behavior. Send something the agent shouldn't handle and confirm it responds appropriately—ideally by explaining what it can and cannot help with and offering to transfer to a human.

When you click Activate without thorough testing, you're debugging in production. Test until you're bored.

---

Step 6: Activate and Deploy to a Channel

Once you're satisfied with testing, click Activate in the builder. Activating the agent makes it available to deploy. It doesn't automatically surface it to end users.

Deploying to Experience Cloud

If you're targeting an Experience Cloud site:

1. Go to the Channels section in your agent configuration.
2. Select Experience Cloud as the channel.
3. Choose the specific site you want to deploy the agent to.
4. In Experience Builder for that site, add the Agentforce chat component to your pages.

The agent will appear as a chat widget on your Experience Cloud site. Visitors interact with it directly.

Deploying to Messaging for In-App and Web

For a website outside of Experience Cloud:

1. Configure a Messaging channel in Setup under Messaging Settings.
2. Link your activated agent to that Messaging channel.
3. Copy the embed code snippet and add it to your external website's HTML.

In both cases, you'll need to configure the agent's End User Access—which users or guest profiles can interact with the agent, and what data the agent can access on their behalf.

---

Step 7: The Conversational Setup Feature (New in Spring '26)

Spring '26 introduced Setup with Agentforce, currently in Beta. This is a separate feature from Agentforce Builder, but it's worth knowing about if you're an admin who spends significant time in Setup.

A chat-style prompt bar appears on the Setup Home page. You type natural language commands—"Create a custom object for project tracking" or "Explain what this permission set gives access to"—and an AI assistant responds with guidance or proposes configuration changes for you to confirm.

What it can help with:

• Creating objects, fields, and relationships
• Navigating to the right Setup page for a specific task
• Explaining existing configuration and formulas
• Troubleshooting common Setup errors

What it does not do:

• Make changes without your confirmation. Every proposed action requires explicit approval before anything is applied.
• Replace manual setup entirely. Complex configurations—particularly permission set assignments, sharing rules, and integration settings—still benefit from deliberate manual review.

The feature is Beta in Spring '26, which means it's available in production orgs but functionality may be incomplete and Salesforce is actively refining it based on feedback. Treat it as a useful accelerator, not a replacement for understanding what you're configuring.

To find it: Open Setup and look for the conversational prompt area on the Setup Home page. If you don't see it, your org may need a system admin to enable it under Einstein/Agentforce settings.

---

Common Gotchas and Limitations

Topic classification is probabilistic. The agent picks the best-matching Topic, not a guaranteed exact match. If two Topics have similar classification descriptions, you'll get misrouting. Keep descriptions distinct and test with real phrasing.

Instructions are not infallible. Large Language Models follow instructions well but not perfectly. An agent with an instruction saying "never discuss pricing" will occasionally discuss pricing if a clever user phrases the question indirectly. Plan for human escalation paths for anything sensitive.

Standard Actions have their own prerequisites. Some out-of-the-box actions require specific Salesforce features to be active. The "Search Knowledge" action, for example, requires Salesforce Knowledge to be set up in your org. Check action documentation before planning your agent's capabilities around one.

Agents can't access data they aren't shared to. The agent acts on behalf of the connected user or guest profile. If your org's sharing rules don't allow the agent's running user to see certain records, the agent can't retrieve them—even if the instruction says to. This is by design (security model is enforced), but it surprises admins building their first agent.

Spring '26 Builder features are partially Beta. The new Agentforce Builder interface with Canvas Mode is itself in Beta as of this release. Expect occasional rough edges and check Salesforce Known Issues before reporting bugs to your team.

Conversation history is session-based. By default, the agent does not retain memory across separate conversations. Each session starts fresh. If your use case requires continuity across sessions, that requires additional configuration.

---

Next Steps

Building your first agent is the easy part. The real work is refining it based on real conversation data — reading what users actually typed, finding the gaps in your Topics, and tightening your instructions over time.

A few resources to bookmark:

• Search Trailhead for "Agentforce Agent Builder" to find the official learning trail with a dedicated practice org
• The Spring '26 release notes have the full changelog for Agentforce Builder, including known limitations for Beta features
• Before deploying to production, confirm your exact licensing model with your Salesforce account team

Keep reading:

• What Is Agentforce 2dx? Salesforce's Proactive AI Explained — where the Agentforce platform is heading after you've built your first agent
• AgentExchange: Top Pre-Built AI Agents for Salesforce in 2026 — pre-configured agents you can deploy without building Topics and Actions from scratch
• Agentforce for Contact Centers: What It Is and Who Should Care — channel-specific configuration for service teams deploying at scale
• Salesforce Spring '26 Admin Guide — the full Spring '26 feature roundup beyond Agentforce Builder