Agentforce League : Session 05 (New Agentforce Builder, Agentforce Voice) - 10th Jan 2026
The fourth session of the AgentForce League 20/26 introduced game-changing features that are transforming how we build and interact with AI agents. With participants from across India, the webinar covered Financial Services Cloud, the revolutionary new AgentForce Builder with AgentScript, and AgentForce Voice capabilities.
Financial Services Cloud: Banking Made Smarter
Financial Services Cloud is Salesforce’s comprehensive solution designed specifically for banks and financial institutions. It handles everything from your everyday savings account to complex loan processing and wealth management.
What makes it special?
Retail Banking: Manages individual customer accounts, cards, and common requests like address updates or freezing credit cards
Wealth Management: Helps financial advisors track customer goals and suggest personalized investment strategies
Digital Lending: Streamlines loan applications with impressive results—one customer reduced loan processing time from 2 days to just 30 minutes!
Pre-built Workflows: Includes ready-to-use templates for transaction disputes, card freezing, and KYC (Know Your Customer) compliance
The New AgentForce Builder & AgentScript: A Game-Changer
This was the star of the session, and for good reason. The new AgentForce Builder with AgentScript represents a fundamental shift in how we create AI agents.
A Crucial Interface Change
Important: The new builder has a completely different layout that can be confusing at first. What looks like the conversation preview window is actually the AgentForce Assistant—an AI helper that guides you through building your agent. As the presenter emphasized: “Mind it, this is not your conversation preview. This is not where you’re going to test the agent. This is the place where you’re going to talk to the agent for helping you.”
The actual testing happens in a separate Preview tab. Don’t mix these up—the AgentForce Assistant is for creating agents, while the Preview tab is for testing them.
The Problem AgentScript Solves
Traditional agents are “probabilistic“—you write prompts like “always do this” or “never do that,” but you’re never quite sure how the agent will interpret them. AgentScript makes agents more “deterministic,” giving you greater control and confidence in how they behave.
What is AgentScript?
Think of AgentScript as a lightweight language that combines the best of both worlds:
You can still write in plain English (declarative)
But you can also add logic, conditions, and procedures (programmatic)
Everything is organized in simple name-value pairs: name: value
The beauty? You don’t need to be a developer to use it.
Three Ways to Build Agents
- Conversational Approach: Simply talk to an AI assistant that helps you build your agent. During the demo, the presenter created a banking agent just by having a conversation:
“I want to create a simple banking agent”
“What should it do?” → “Check account balance”
“Any actions needed?” → “Retrieve current account balance”
The system automatically generated all the code!
- Hybrid Approach: Write natural language instructions but embed expressions and logic where needed.
- Full Script Mode: Write complete AgentScript code for maximum control.
Understanding AgentScript Blocks
AgentScript organizes everything into easy-to-understand sections called “blocks”:
System Block: Welcome messages, error messages
Config Block: Agent name, settings
Variables Block: Store information the agent needs
Mutable variables: Can be changed by the agent
Linked variables: Only store results from processes
Start Agent Block: Where your agent begins
Topics Block: The actual conversations and logic
What's New and Powerful
Before Topic Logic: Run pre-processing before the agent even starts thinking. For example, fetch account information before the conversation begins.
After Topic Logic: Do post-processing after getting the AI’s response.
This “separation of concerns” makes agents faster because they’re doing less thinking and more doing.
Live Demo Highlights
The presenter demonstrated creating a “Hello World” agent that greets users with a poem. The magic happened in the Preview section, where powerful debugging features were revealed:
Revolutionary Debugging Tools:
Timeline View: See exactly how long each step took (in seconds)
Summary View: Get a plain-English explanation of what the agent did
Detailed Inspection: Click on any step to see the exact input, topic selection, and reasoning
Error Highlighting: Failed steps appear in red for easy troubleshooting
Inside the New AgentForce Builder
Important Technical Notes
Beta Status: The new builder is actively being developed. A patch was deployed during the session week, showing how rapidly it’s evolving
Metadata Type: Agents built with AgentScript use AiAuthoringBundle (different from the old GenAiPlannerBundle)
Access: Launch from App Launcher → search for “AgentForce Studio” (not from Setup)
AgentForce Voice: Bringing Agents to Life
AgentForce Voice isn’t a separate product—it’s simply another channel where you can deploy your existing agents. Think of it as giving your agent the ability to talk on the phone.
Why Voice Matters
Statistics reveal that 85% of customer interactions happen via phone, not chat. This makes voice capabilities essential for complete customer service automation.
Two Voice Channels
Phone Channel (Generally Available): Agents answer incoming phone calls with ultra-realistic voices
Digital Channel (Coming Soon): Add a microphone button to website chat interfaces
What Makes It Different from IVR
- Traditional IVR systems are static: “Press 1 for billing, press 2 for support.”
- AgentForce Voice is an Interactive Voice Agent (IVA) that:
- Understands natural language (no menu navigation needed)
- Sounds almost human (not robotic)
- Can be interrupted mid-sentence (just like a real conversation)
- Uses the same reasoning engine as text-based agents
How It Works Behind the Scenes
- Speech to Text: Your voice is converted to text using specialized AI
- Reasoning: The agent processes your request using its standard logic
- Text to Speech: The response is converted back to natural-sounding voice
AgentForce Voice vs. Service Cloud Voice
This distinction is crucial:
AgentForce Voice:
- Answers calls BEFORE humans get involved
- The first line of defense for incoming calls
- Tries to resolve issues autonomously
- Transfers to humans only when needed
Service Cloud Voice:
- Used BY human agents to take calls
- Provides features like automatic transcription
- Suggests next best actions based on conversation keywords
- Works AFTER AgentForce Voice transfers the call
Think of it this way: AgentForce Voice is the automated receptionist; Service Cloud Voice is the tool human agents use once they pick up the phone.
Resources and Helpful Links
Author: Tejas Jain
