Introduction to Salesforce Agent Script: Build Predictable AI Agents

Agent Script solves this problem. It’s a declarative scripting language in Salesforce Agentforce that gives you precise control over how your agents think, respond, and act. Instead of leaving everything to an LLM, Agent Script lets you define clear logic, structured workflows, and topic transitions – ensuring your agents behave consistently and deliver reliable outcomes.

This combination of natural language readability and structured logic makes Agent Script a must-have skill for anyone building intelligent, business-aligned workflows in Salesforce. In the sections ahead, we’ll explore what Agent Script is, how it works, and why it’s transforming the way teams design AI-powered experiences.

Agent Script is the language you use to build smart, reliable agents in Agentforce Builder. It blends the ease of natural language with the structure of logic-based expressions. With Agent Script, you can set up clear rules like if/else conditions, define how an agent should move between topics, update or compare variables, and decide which actions to run.

It introduces a “hybrid reasoning” approach that blends the creative flexibility of Large Language Models (LLMs) with the precision of structured business logic. This makes your agent predictable and context-aware instead of relying only on an LLM to interpret everything. For example, you can decide exactly when the agent should switch topics or run a series of actions in a specific order. This level of control helps you create smoother, more consistent workflows that align with real business processes.

Agent Script is a specialized language built by Salesforce for defining how agents behave in Agentforce. It combines both declarative and procedural elements so you can express what an agent should do and how it should do it, step by step. When you save your script, it’s compiled into metadata that the reasoning engine uses to run the agent. The language is designed to be readable, even for non-developers, and relies on structured blocks and indentation to show relationships.

Agent Script uses the @ symbol to quickly reference key elements in your agent -such as actions, topics, variables, and outputs – without writing long or complex expressions. This keeps your script clean and readable while making it easy to connect logic to the right resources. You simply point to what you need by using the appropriate prefix:

This pattern creates a consistent and intuitive way to work with your agent’s building blocks, helping you move faster and avoid errors.

Agent Script is organized around a property-based format where each line follows a key: value pattern. The key always appears before the colon, and the value comes after it. description: “Get account info”

Agent Script uses indentation to define structure, so formatting must stay consistent. You can use either two spaces or a tab, but you must choose one style and apply it throughout the script to avoid errors. Indentation shows which values belong to a property, and the arrow symbol (->) is used to introduce logic steps. For multiline messages, the pipe (|) lets you write text across several lines or switch into a prompt-style instruction within your logic

Agent Script lets you add simple logic to your workflows using familiar expressions. You can control decisions with if and else, do basic math with + and -, and compare values using ==, !=, >, and <.
It also supports checking for empty or missing data with is None and is not None. These tools help your agent respond correctly based on the information it receives.

Agent Script lets you add comments using the # symbol. Anything written after # on the same line is ignored during execution. This is an easy way to document your logic, explain decisions, and help others understand the script when they read or update it later.

Agent Script organizes logic into blocks, each serving a specific purpose in defining how your agent behaves. At a high level, blocks group related settings, data, and procedures to make scripts clear and modular. Core Block Types

Each block type makes scripts easier to structure and understand, helping you define both data and behavior in a clear, maintainable way.

This flow keeps conversations consistent and easy to manage.

Appointment Scheduling Topic (Agent Script vs. Natural Language Instructions)

Before Agent Script, we just used plain text to tell agents what to do. It was easy to set up, but it didn’t give developers enough fine-grained control over the actual steps the agent took.

New Approach – Agent Script

As AI becomes an integral part of business operations, predictability and control are essential for delivering consistent customer experiences. Salesforce Agent Script addresses this need by combining the flexibility of natural language with the precision of structured logic. Its hybrid reasoning approach ensures agents can handle complex workflows while remaining conversational and adaptive. By leveraging clear syntax, topic-based organization, and intuitive references, teams can design intelligent agents that align with business rules and deliver reliable outcomes.

https://developer.salesforce.com/sample-apps/agent-script-recipes/getting-started/overview

https://github.com/trailheadapps/agent-script-recipes

https://developer.salesforce.com/docs/einstein/genai/guide/agent-script.html

https://developer.salesforce.com/docs/einstein/genai/guide/ascript-reference.html

https://www.salesforce.com/agentforce/script/

https://trailhead.salesforce.com/users/pozil/trailmixes/agent-script