Agentforce Builder

Agentforce Builder is the visual (low-code) environment Salesforce offers to create, configure and publish AI agents on its platform. Instead of programming the agent's logic line by line, the builder lets you declaratively define the subagents the agent knows how to handle (Salesforce used to call them topics), the instructions that guide its behavior and the actions it can execute (for example, looking up a case, creating an opportunity or triggering a flow).

The key point is that it separates the agent's design from its reasoning engine: the builder describes what the agent can do and how it should behave, while the AI model decides when and in what order to use those capabilities. It is the layer where a business team or a Salesforce Administrator builds a useful assistant without depending on a developer for every change.

In projects that extend the platform with custom development, this configuration lives alongside Apex code and custom components; it is part of the work covered by the Salesforce Developers team when an agent needs actions that the declarative builder does not provide out of the box.

How it works in practice

Building an agent in Agentforce Builder follows a repeatable pattern. First you define the agent's role and tone (who it serves and how it should express itself). Then you load the subagents, which are groupings of problems the agent knows how to solve: "check an order's status," "schedule a visit," "answer billing questions." Each subagent has instructions in natural language and a set of associated actions.

Actions are the agent's real muscle. They can be a Salesforce Flow, an Apex class, a reusable prompt or a call to an external API. The builder exposes them as blocks that the agent invokes through tool calling, and the reasoning engine (the Atlas Reasoning Engine) chooses which one to use based on the conversation. Before publishing, there is a testing panel where you simulate the conversation and see, step by step, which subagents and actions were triggered.

Why it matters for a business

The value lies in the speed of iteration and in who can operate it. Without a visual builder, every adjustment to a conversational assistant requires touching code and deploying. With Agentforce Builder, an operations team or an administrator can add a subagent, fix an instruction or add an action and republish in minutes, keeping version control and testing within the same platform. This lowers the barrier for an AI agent to evolve at the pace of the business.

A concrete example (Argentina)

A consumer goods distributor in Buenos Aires wants its customer service agent, connected to WhatsApp, to answer order status questions and hand off to a human when the customer asks for a credit note. In the builder, the team defines an "Order tracking" subagent with an action that queries the dispatch system, and another subagent, "Billing claims," whose instruction is to always escalate to a human agent. When the returns policy changes, no software release is needed: the subagent's instruction is edited in the builder and republished.

Common mistakes when using it

• Subagents that are too broad or overlap: if two subagents cover the same thing, the agent gets confused about which one to trigger. Each subagent should have a clear, narrow scope.
• Vague instructions: saying "help the customer" is not enough; precise instructions (what to do, what never to do, when to escalate) are what separate a reliable agent from an erratic one.
• Forgetting the human handoff: every agent needs a way out to a person for the cases it should not resolve on its own.
• Not testing with real cases: the testing panel exists to detect, before production, the paths where the agent hallucinates or picks the wrong action.

How it differs from Agentforce as a whole

It is common to confuse the tool with the platform. Agentforce is Salesforce's complete AI agent product; Agentforce Builder is the design surface within that product. The following table sorts out the related terms:

In short, the builder is the front door for building an agent without code, while custom actions, versioning and complex deployments move into the territory of Agentforce DX and development on the platform.