Agentforce Updates & Release Notes

WHAT'S NEW IN AGENTFORCE

Agentforce is ready for bigger jobs. Launch purpose-built agents faster, keep complex work moving across days and week, and continuously improve performance — all with humans in control.

EXPLORE WHAT'S NEW

Five new ways to put agents to work

PURPOSE-BUILT AGENTS

Start ready for the job

Launch agents for sales, service, HR, and more, with job-specific subagents, prebuilt actions, and Salesforce context already in place.

Explore purpose-built agents

Take a closer look

Built for real work, from day one

Explore how each capability helps teams launch faster, stay productive longer, and get more value from every agent.

Portfolio of Purpose-Built Agents

Start ready

Move past the blank canvas and put proven agent patterns to work immediately. Each purpose-built agent is designed for a specific role across sales, service, HR, supply chain, and more, with the subagents, actions, and Salesforce data context needed to begin delivering value. Teams can spend less time assembling the basics and more time tailoring the experience to the way their business works.

Learn more

Long-Horizon Runtime

Keep work moving

Give agents goals that extend beyond a single conversation. The Long-Horizon Runtime maintains context across days and weeks, adjusts the plan as conditions change, and continues coordinating follow-ups across a multi-touch workflow. Built-in checkpoints keep people informed and in control, so agents can handle longer-running work without turning autonomy into a black box.

Learn more

Agent Optimizer

Improve every agent

Turn agent improvement into an ongoing practice. Agent Optimizer observes performance, surfaces friction in instructions and workflows, and recommends specific changes teams can act on. Because those insights appear inside the existing workflow, builders can move from identifying a problem to tuning the agent faster — and keep improving quality as real-world usage grows.

AI Skills in Coworker

Teach once. Work anywhere.

Capture a useful task once, then make it available to AI teammates across the places people already work. AI Skills can surface in Slack, Microsoft Teams, desktop, mobile, ChatGPT, and Claude, helping teams reuse trusted ways of working instead of rebuilding the same capability channel by channel. The result is a more consistent experience wherever work begins.

Multi-Agent Orchestration

Turn agents into teams

Bring specialized agents together around a shared outcome. Multi-Agent Orchestration dynamically sends work to the right agent, carries context between each step, and coordinates handoffs across systems and departments. Customers and employees get one connected journey while every agent contributes the expertise it was built to provide.

Learn more

Ready to take the next step with Agentforce?

Build agents fast.

Take a closer look at how agent building works in our library.

Watch demos

Get expert guidance.

Launch Agentforce with speed, confidence, and ROI you can measure.

See how

Talk to a rep.

Tell us about your business needs, and we’ll help you find answers.

Contact us

Original source

Bring the power of assistive and autonomous AI to your business with Agentforce.

• Agents Now Default to Error Mode for File Uploads
  Agentforce agents that can receive file uploads from a client now default to error mode, so you have more control over agent file handling. Previously, agents that can receive files defaulted to auto mode, and your agent decided how to handle the file. In error mode, when someone uploads a file, the agent sends a customizable error message. If you want your agent to accept file uploads, you must add the file upload parameter to your agent’s script and specify a mode. If you don't want your agent to accept file uploads but you don't want it to send an error message, you can specify the disabled mode in the file upload parameter of your agent's script. You can also control which clients can accept file uploads.

• GPT-6 Luna, Sol, and Astra Are Generally Available
  The GPT-6 Luna, Sol, and Astra models are ready for use in production with a rate limit of 1,000 requests per minute. You can use these models in any supported feature, such as in a prompt template.

• Claude Opus 5.5 Is Generally Available
  The Claude Opus 5.5 model is ready for use in production with a rate limit of 1,000 requests per minute. You can use this model in any supported feature, such as in a prompt template.

• Claude Opus 5 Is Generally Available
  The Claude Opus 5 model is ready for use in production with a rate limit of 1,000 requests per minute. You can use this model in any supported feature, such as in a prompt template.

• Free and Trial Orgs Have New Rate Limits and Model Access Rules
  These changes may affect Developer Edition (DE) Orgs, Trailhead Playgrounds, and more.

• Ground Prompt Templates with Microsoft Office, Text, and Email Files
  Prompt templates in Prompt Builder can ground generative AI requests with Word documents, spreadsheets, presentations, text, CSV files, and email files, in addition to PDF and image files.

• Conditional Logic in Prompt Templates Is Generally Available
  Conditional logic in generative AI prompt templates is generally available. Add if, else if, and else blocks so that one prompt template adapts based on grounded data.

• Format Prompt Instructions with Text Styling in Prompt Builder
  The Block Editor now supports rich text formatting in generative AI prompt templates. Headings, lists, bold, italic, underline, and dividers are preserved through save, Preview, and generative AI at run time.

• See Knowledge Sources in Agentforce Session Details
  See Salesforce Knowledge articles and external sources your agent used in a session. Open a session from Sessions & Intents to review which sources were used to support the agent’s responses.

• Track Voice Latency and Speech Metrics in Session Trace
  Spot long Agentforce Voice silences and inspect speech-to-text and text-to-speech timing in Session Trace, including word-level automatic speech recognition (ASR) confidence. A clock icon flags abnormal agent silence.

• Review Updated Agentforce Observability Metrics
  Several Agentforce Observability metric names and formulas changed. Compare latency and completion trends with the new labels, and review Success Rate with the updated calculation. Agent Interaction Duration is removed from dashboard templates.

• Legacy Agent Builder Changing Reasoning Engine from GPT-4o to GPT-5.6 Luna
  The legacy Agentforce Builder is implementing an on/off switch to change the reasoning engine to GPT-5.6 Luna on October 15, 2026.

• Dynamic Default Model Selection for New Agents
  Starting October 15, 2026, new agents created in Agentforce Builder default to an optimal model selected for them.

• Authenticate to External MCP Servers with Per-User Credentials
  Allow Agentforce Employee agents (AEA) to act on behalf of individual users with user-based Model Context Protocol (MCP) servers. With per-user authentication, your users can access an external MCP server with their own credentials instead of a shared system-to-system connection. Eliminate shared service-user tokens and ensure a clear per-user audit trail.

• Gemini 3.8 Flash Is Generally Available
  The Gemini 3.8 Flash model is ready for use in production with a rate limit of 1,000 requests per minute. You can use this model in any supported feature, such as in a prompt template.

• Spend Less Time on Manual Testing with the Agentforce Assistant (Beta)
  The Agentforce assistant no longer just suggests changes to your agent. Now it can test them for you, too. When the assistant suggests a change, it automatically runs previews in simulate mode to validate your agent’s basic functionality and its performance against variations and edge cases. Then it iterates until your agent behaves as expected, so you can accept the assistant’s changes with less manual verification. You can also prompt the assistant to test your own changes. For example, ask it to run a specific subagent through different personas and refine based on the results.

• Build and Activate Agents with a Simpler Versioning Process
  You no longer have to manually commit an agent, simplifying how you edit, version, and activate agents. Previously, you committed your agent version before you could activate it. Now, you can activate a draft version of the agent, and Agentforce Builder commits it in the same step. The Commit Version button has been removed.

• Azure OpenAI o3 Family and o4-mini Reroute Date Approaching
  Azure OpenAI o3 family and o4-mini model requests will be rerouted November 11, 2026.

• Upgrade Einstein Bots to Agentforce Service Agent (Generally Available)
  The Upgrade Agent tool, previously available in beta, is now generally available. Use the tool to convert an Einstein Bot to an Agentforce Service Agent. The tool creates the new agent separately, so your existing bot remains available while you review and configure the new agent.

• Let Agentforce Agents Remember What Matters to Your Users (Generally Available)
  With Agent Memory, your Agentforce agents recall relevant details from your users' past conversations, so your users don't have to repeat themselves. Agents also reference profile details, such as a user's name, from the first message, so responses feel personalized from the start. Previously, agents had no memory of past conversations, so your users had to restate the same details every time they started a new conversation with an agent.

• Preview Custom Lightning Types in Agentforce Builder with Fewer Steps
  Preview Custom Lightning types (CLTs) in Agentforce Builder for actions you create or customize directly in an agent. CLTs are used to customize the UI for agent action inputs and outputs for a more branded, interactive experience. Previously, CLTs didn't render in preview for actions created or customized in an agent in Agentforce Builder. You had to recreate the action in the asset library. You also had to commit your agent before CLTs rendered at all. Now, CLTs render for actions in preview no matter where you created or modified them, or whether your agent is in a draft or committed state.

• Learn How to Optimize an Agent by Analyzing Production Data
  We added documentation that shows you how to headlessly analyze production data to identify suboptimal agent responses. Your AI assistant identifies root causes, turns them into durable test cases, and suggests fixes, which you validate against a full test suite in your testing environment, such as a sandbox, before promoting to production.

• Learn How to Ground an Agent with an Agentforce Data Library
  We added documentation that shows you how to develop an agent, ground it with an Agentforce data library, test, and activate it. This example shows you how to complete the steps by using headless (using an AI coding assistant), or Canvas (using the Salesforce UI) paths.

• Learn How to Build and Deploy an Enhanced Chat Agent
  We added documentation that shows you how to build a chat agent, test and activate it, and deploy it to an Enhanced Chat v2 channel. This example shows you how to complete the steps by using headless (using an AI coding assistant), Canvas (using the Salesforce UI), or Agent Script (using code) paths.

• GPT-5 Family (GPT-5, GPT-5-Mini, GPT-5-Nano, GPT-5-Pro) and O3 / O3-Pro Reroute Date Approaching
  GPT-5 Family and O3 / O3-Pro model requests will be rerouted December 11, 2026.

• Help Customers Continue Ended Sessions in Enhanced Chat v2
  After a messaging session times out or ends in Enhanced Chat v2, customers can easily resume it by clicking Continue Conversation in the chat window.

• Give Customers a Smoother Conversation Experience in Enhanced Chat v2
  Enhanced Chat v2 now delivers a smoother, easier-to-follow conversation experience. When a customer sends a message, it stays in view while the response loads instead of scrolling out of sight, so customers can read replies in context. The avatar now appears above the Send Message text field for a cleaner layout.

• Give AI Agents More Context with File Uploads
  Agentforce Service and Employee agents now support file uploads in production and preview conversations, so you can help your agents resolve more inquiries. When you add file upload support to your agent, your users can share screenshots, receipts, and PDFs by attaching a file in the chat window. Your agent can interpret the image or PDF content and answer questions, and it can add the file to a Salesforce record or use the file as context. File uploads are supported for agents connected to web clients on Enhanced Chat v1, Enhanced Chat v2, and Lightning Experience.

• Control How Agent Action Outputs Appear on a Connection
  Always format an agent action's output the same way on a specific connection with the render nested inline action. If you don't specify an adaptive response format, Agentforce decides which one to use during the conversation. You can also format an action's output as JSON so that a connection can receive the action's structured data directly. And, with show_and_return nested inline action, you can now tell the agent to send a response and end reasoning at the end of an action.

• Customize Agent Behavior Across Channels with Connection Instructions
  Guide how your agent reasons and responds on the Messaging, Service Email, or Marketing Email connection with customizable connection instructions. For example, you can tell the agent to ignore disclaimers in emails or to use a specific response format for certain messages.

• Customize Agent Responses with Dynamic Messaging Components
  Create custom response formats based on messaging components, including the enhanced link, question with static options, question with dynamic options, time selector, and form components. Use the output of an agent action to populate dynamic content in your custom response formats. For example, have an agent action gather the details for a return request and then populate a custom form for the customer.

• Easily Compare and Merge Agent Versions in Agentforce Builder
  Copy improvements from one agent version to another with a few clicks. You can now compare two versions and merge changes from one into another in Canvas and Script view.

• Keep Field Generation Responses Focused on Business Use Cases
  By default, field generation follow-up refinement requests aren't validated to ensure relevance to business requests. This means users can make unrelated, non-business requests within the field generation panel. To steer users back to your specific business goals, turn on the new Validate Field Generation Refinements setting in Setup. This limits the agent's focus entirely to field generation-related activity, ensuring your Einstein credits are spent only on relevant business use cases.

• Field Generation Refinements Now Respect Your Selected Model
  Previously, follow-up responses from Einstein in the field generation panel used the Salesforce default LLM instead of the model assigned to your field generation prompt template. Follow-up and refinement responses are now generated from the model that you selected in the prompt template.

• Gemini 3.7 Flash Is Generally Available
  The Gemini 3.7 Flash model is ready for use in production with a rate limit of 1,000 requests per minute. Use these models for certain functionality, such as in a prompt template.

• Gemini 3.6 Flash Is Generally Available
  The Gemini 3.6 Flash model is ready for use in production with a rate limit of 1,000 requests per minute. Use these models for certain functionality, such as in a prompt template.

• Gemini 2.5 Pro, Flash, and Flash-Lite Reroute Date Approaching
  Gemini 2.5 Pro, Flash, and Flash-Lite model requests will be rerouted October 20, 2026. Gemini 2.5 Pro and 2.5 Flash will be rerouted to Gemini 3.5 Flash, and Gemini 2.5 Flash-Lite will be rerouted to Gemini 3.5 Flash-Lite. We recommend that you test your prompts and applications with the new model in Prompt Builder and AI Models as soon as possible, because expected responses can change.

• Move Agents Between Orgs More Easily with Simplified Agent Metadata Types
  Starting in API version 68.0, we’ve streamlined Agentforce agent metadata types to make it easier to retrieve and deploy agents. Previously, moving an agent from sandbox to production required at least three metadata types. Plus, you had to manually add every Apex class, flow, and prompt template associated with an agent action, making sure to indicate the correct agent version. With the new AiAgentDefinition and AiAgentDefinitionVersion types, Salesforce CLI now retrieves the agent and all its dependencies automatically.

• Agentforce Platform Enabled by Default
  In Winter ’27, the Agentforce platform is enabled by default for all orgs with Agentforce access, as granted by SKUs, licensing, and editions. New orgs are enabled at creation, and existing orgs will be auto-enabled. The Agentforce setting will be removed from the Agentforce Agents page in Setup, so you can build, test, and deploy agents with one less step. The platform is enabled at no additional cost and there are no changes to billing.

• Global Expansion for Agentforce Voice with Dynamic Routing
  To expand the reach of Dynamic Routing, we are continuously adding local phone numbers in more countries throughout this release.

• Dynamic Routing for Agentforce Voice Is Now Available in Japan
  Dynamic Routing now supports local phone numbers in Japan. To connect to the voice-enabled agent, call the reserveRoutableNumber API, and then forward the call to the local number.

• Show Full Conversation History in Post-Call Transcripts for Agentforce Voice Calls
  The Enhance Conversation component in the Service Console now displays consistent transcripts during and after a voice call handled by a voice-enabled agent. When the agent transfers the call to a service rep, the voice call record for the rep-customer conversation (VC1) now shows both the agent-customer and rep-customer transcripts after the call, matching what was visible during the call. Previously, VC1 showed both transcripts during the call but dropped the agent-customer transcript after the call ended.

• Agentforce Voice Now Enforces the Limit on Procured Phone Numbers
  Agentforce Voice now enforces the Salesforce Voice (Native Telephony) limit on the number of phone numbers that you can procure. In production orgs, your Agentforce Voice license determines the limit. In sandbox and trial orgs, the limit is 10 phone numbers regardless of your license. Previously, these limits weren't enforced.

• Pass Context with Custom SIP Headers and Disconnect with REFER in Agentforce Voice
  Capture custom SIP X-headers from inbound calls to deliver contextual information, such as customer intent, language preference, or case history. Configure your escalation and fallback flows to return or modify headers using the SIP REFER method when disconnecting or transferring calls.

• Extend Agentforce Solutions with Multi-Agent Orchestration (Generally Available)
  Extend your agent’s capabilities with Multi-Agent Orchestration for Agentforce. Connect your Agentforce agent with other specialized Agentforce agents in your Salesforce org. Together they collaborate on complex tasks to deliver amazing outcomes for your business.

Original source

This release is in preview

Note: This release is in preview. Features described in this document don't become generally available until the latest general availability date that Salesforce announces for this release. Before then, and where features are noted as beta, pilot, or developer preview, we can't guarantee general availability within any particular time frame or at all. Make your purchase decisions only on the basis of generally available products and features.

Bring the power of assistive and autonomous AI to your business with Agentforce.

• Gemini 2.5 Pro, Flash, and Flash-Lite Reroute Date Approaching

  Gemini 2.5 Pro, Flash, and Flash-Lite model requests will be rerouted to Gemini 3.5 Pro, Flash, and Flash-Lite on October 20, 2026. We recommend that you test your prompts and applications with the new model in Prompt Builder and Einstein Studio as soon as possible, because expected responses can change.

• Agentforce Platform Enabled by Default

  In Winter ’27, the Agentforce platform is enabled by default for all orgs with Agentforce access, as granted by SKUs, licensing, and editions. New orgs are enabled at creation, and existing orgs will be auto-enabled. The Agentforce setting will be removed from the Agentforce Agents page in Setup, so you can build, test, and deploy agents with one less step. The platform is enabled at no additional cost and there are no changes to billing.

• Show Full Conversation History in Post-Call Transcripts for Agentforce Voice Calls

  The Enhance Conversation component in the Service Console now displays consistent transcripts during and after a voice call handled by a voice-enabled agent. When the agent transfers the call to a service rep, the voice call record for the rep-customer conversation (VC1) now shows both the agent-customer and rep-customer transcripts after the call, matching what was visible during the call. Previously, VC1 showed both transcripts during the call but dropped the agent-customer transcript after the call ended.

• Agentforce Voice Now Enforces the Limit on Procured Phone Numbers

  Agentforce Voice now enforces the Salesforce Voice (Native Telephony) limit on the number of phone numbers that you can procure. In production orgs, your Agentforce Voice license determines the limit. In sandbox and trial orgs, the limit is 10 phone numbers regardless of your license. Previously, these limits weren't enforced.

• Enable Flexible Call Recording Control Within Conversations via Toolkit API

  Salesforce developers and admins can now use the Salesforce Voice Toolkit API to pause and resume call recordings without the Control Call Recording user permission. Previously, these users needed this permission. This permission is still required to use the pause and resume controls in the Salesforce UI.

Original source

Bring the power of assistive and autonomous AI to your business with Agentforce.

• Support More Users with Enhanced Chat v2 Accessibility Improvements

Reach more users with a more accessible Enhanced Chat v2 experience. We’ve improved keyboard focus management, accessible names for controls, screen reader citations, and mobile navigation on Android. Enhanced Chat continues to align with WCAG 2.2 Level AA and you can track resolutions through the Voluntary Product Accessibility Template (VPAT).

• Extend Agentforce Solutions with Multi-Agent Orchestration (Generally Available)

Extend your agent’s capabilities with Multi-Agent Orchestration for Agentforce. Connect your Agentforce agent with other specialized Agentforce agents in your Salesforce org. Together they collaborate on complex tasks to deliver amazing outcomes for your business.

• Easily Resolve Out-of-Sync MCP Server Registrations

MCP server providers own their servers and can change server and tool settings at any time. Server-side changes make your current registration settings out of sync, which can adversely affect agents that rely on those MCP server tools. Agentforce now monitors and identifies these discrepancies. It flags a registered MCP server or tool as Out of Sync, and notifies an admin to take corrective action.

• Add Tools from First-Party MCP Servers to Your Agents

AAgents can now use tools from Salesforce-hosted (first-party) MCP servers. First-party tools run in the context of the Salesforce user through an internal credential. Each agent calls the MCP tools with its own user's permissions. Previously, agents could get tools only from third-party MCP servers accessed through a Named Credential.

• Investigate Agents Faster with Streamed Session Trace Data

Agentforce Session Tracing data now arrives in Data 360 through streaming ingestion. The result? Wait time reduced from 15–20 minutes to as little as 30 seconds to 2 minutes. Alert notifications for monitored metrics arrive much more quickly. Resolve emerging agent issues sooner, before they become customer-facing incidents.

• Apply Agentforce Coworker's Personal Context to Grid Worksheet Rows

Turn a list of Salesforce records into a repeatable Agentforce Coworker workflow. Add Agent from the Add Column menu and describe the work you want done, such as “prepare these accounts for tomorrow’s meetings.” Agentforce Coworker reviews your relevant context, including account relationships, calendar information, and upcoming meetings. It then applies the instruction across your worksheet and writes a tailored result for each record, so you can review, compare, and act on the complete set in one place.

• Read Agent Column Outputs as Formatted Text in Grid

The output preview for Agent columns in Agentforce Grid shows formatted text instead of raw JSON, so results are easier to read at a glance. A JSON tab is still available in the preview for cases where you want to inspect the raw output. Previously, Agent column previews showed raw JSON by default, requiring manual parsing to understand the results.

• Evaluate Entire Conversations for Voice and Chat Agents with Testing Center in Agentforce Studio (Beta)

Simulate and score full multi-turn conversations at scale to validate that your voice and chat agents resolve customer issues end-to-end. Generate up to 20 cases or upload up to 1,000 test cases, run them against realistic customer personas, and get scored results with transcripts and audio recordings.

• Deploy Agentforce Voice in Japanese in Your Production Environment

Deploy Agentforce Voice in Japanese, now generally available for production environments. Set Japanese as your agent's default language and use a verified Japanese voice to serve customers on live calls.

• Analyze Sessions with Percentile Latency, Feedback, and Connected-Agent Data

Pinpoint performance issues with percentile latency metrics and connected-agent insights, and understand the usage, knowledge sources, and feedback associated with each session. Reusable views, simplified evaluation workflows, and customizable analytics make it easier to investigate sessions and improve agent performance.

• Write Precise Agent-Level Instructions in Canvas View

Spend less time writing agent-level instructions with a dedicated page in Canvas view. You can also add variables to agent-level instructions and format them with Markdown.

• Understand an Agent at a Glance with an Agent Summary

Quickly review a new agent or confirm that an agent is set up the way you intended before you activate it. Agentforce Builder now generates and shows a visual summary of the agent and its subagents on the first page in Canvas view.

• Adapt Your Agent's Responses to the User's Language (Beta)

Set up your agent faster and give your users a more natural conversational experience. When you turn on adaptive language mode, your agent detects the language of each incoming message and responds in the same language. Your agent can use any language its AI model supports, including languages that aren't in the Allowed Languages list.

• Register and Manage MCP Servers with Headless Tools

Deliver Headless 360 solutions for Model Context Protocol (MCP) for Agentforce. Bring your preferred headless development tools and frameworks to register external MCP servers and manage their allowlisted assets. Shift development from UI-driven interactions to agentic, customer-driven implementations.

• Proactively Monitor Agent Health

Agent Health Monitoring detects silent agent failures—like spiking error rates or high latency—in near-real time across your AI agents. Create alerts for standard or custom metrics, then get notified when thresholds are breached. Use session traces to debug conversation flows and identify root causes. Address issues quickly so you can mitigate customer impact.

• Greet Users with a Chat Invitation in Enhanced Chat v2

Invite users to start a chat with a service rep or AI agent in Enhanced Chat v2. You can use the default invitation or create one with custom text, rules, conditions, and branding on the Embedded Service Deployments Settings Setup page.

• Make Voice Conversations More Engaging in Enhanced Chat v2

Give your users more ways to interact with an agent during a voice conversation in Enhanced Chat v2. A transcript now appears in the chat as your customer and the agent speak, so users can switch between talking and messaging without losing the thread. Users can also minimize the chat window and continue speaking with the agent.

• Deploy Agents to Enhanced Chat v2 Faster in Agentforce Builder

Spend less time setting up Enhanced Chat v2 deployments with the latest improvements in Agentforce Builder. Now, when you create a channel from the Enhanced Chat v2 Settings page, Agentforce creates an embedded service deployment for you. You can also easily add more deployments to a channel from the Enhanced Chat Channels table.

• Spend Less Time Diagnosing Agent Errors with More Actionable Error Messages

During preview conversations, Agentforce Builder shows more detailed error messages for some platform services, replacing errors that were silent or difficult to diagnose. When a service encounters a problem during a preview conversation, you can open the Trace tab to see what went wrong. All error messages include a link to documentation covering possible causes, troubleshooting steps, and guidance on when to contact Salesforce Customer Support. This change doesn't affect end user-facing error messages.

• Take the Guesswork out of Agent Building with Improved AI Assistance (Beta)

We’ve upgraded the embedded AI assistant in Agentforce Builder, so you can focus on what your agent should do instead of how to build it. It asks more targeted follow-up questions to understand your intent and shape a plan, then proposes and explains changes before applying them. When it makes changes, you can trust it to generate valid Agent Script. And when you use it for troubleshooting, it more accurately diagnoses issues and proposes more effective solutions.

• Get Always-On Insights with Agent Observability

For new Salesforce orgs only, turning on Agentforce automatically turns on agent observability features by default. Data collection starts for Agentforce Session Tracing, Audit and Feedback, and related features. The observability user experience is available in Agent Analytics and Agentforce Studio.

• Enhanced Logging for Agent Sessions

Introducing AiAgentSessionLog, a comprehensive logging data model that captures all notable events emitted during AI agent sessions. AiAgentSessionLog provides a unified way to track informational, warning, debug, and error-level events across your AI solutions. Analyze run-time behavior, troubleshoot failures, and trace issues across single-agent and multi-agent flows.

• Let Callers Key In Numbers Instead of Speaking Them

Agentforce Voice now supports keypad input on phone calls. Instead of saying an order number or account number aloud, callers enter it on their phone keypad, which is faster, more accurate, and more private. Callers press the pound key (#) to finish, or the input completes automatically after a short pause.

• Access Agentforce Grid from Agentforce Studio

Agentforce Grid now has a second entry point inside the Agentforce Studio app, so you can find it right alongside the Testing Center and your other Agentforce building tools. Just select Grids from the tools panel. The original App Launcher entry point for Agentforce Grid still works, too.

• Learn Agentforce Grid Basics Right from the All Grids Page

The All Grids page now surfaces Trailhead, Help, and Claude links so you always know where to go for help with getting started. Agentforce Grid gives you an open canvas for building AI-powered workflows, but an empty worksheet doesn't tell you what's possible. Now, a welcome section on the All Grids page offers getting-started resources every time you visit, so you can learn the basics before you start building.

• Assess Security Risk Scores for Third-Party MCP Servers

Check the risk score of third-party Model Context Protocol (MCP) servers during registration. Agentforce Registry conducts an automatic security scan that returns a Low, Medium, or High risk score. Use the risk score as a risk indicator of MCP servers.

• Customize Conversation Recommendations in Agentforce Builder

You can now customize conversation recommendations for Agentforce Employee agents in Canvas view, in addition to Script view. Conversation recommendations refer to both welcome recommendations, which appear on the welcome screen when the Agentforce panel opens, and recommendations that the agent suggests during a conversation. Use them to guide employees to your most relevant and frequently used actions.

• Refine Web Search Results by Using Excluded Domains

Exclude up to 10 domains from Search the Web agent action results to ensure your agents retrieve relevant information. Excluded domains are configured per action and apply regardless of the selected search provider.

• Gain Agent Observability Insights Without Consuming Data 360 Credits

Gain comprehensive agent monitoring and insights with Agent Analytics and related tools—without credit consumption overhead. For agent observability features, data collection, metrics calculations, queries, reports, and dashboards no longer consume Data 360 credits.

• Agentforce Platform Enabled by Default Starting August 2026

In August 2026, we plan to turn on the Agentforce platform for all orgs with Agentforce access. New orgs will be enabled by default. We’ll remove the Agentforce toggle from the Agentforce Agents page in Setup, so you can build, test, and deploy agents with one less step. The platform is provisioned at no additional cost and there are no changes to billing.

• Cover More Cases in a Condition with Else If Branches

Guide your agent down unique paths for different cases in a single condition. Previously, you could only add If and Else branches to a condition statement. Now, you also add one or more Else If branches, and the agent follows the first branch that applies.

• Preview Custom Lightning Types in Agentforce Service Agents

Now you can validate your Agentforce Service agent’s custom responses in Agentforce Builder. Custom Lightning types (CLTs) are used to customize the UI for agent action inputs and outputs for a more branded, interactive experience. When you preview a conversation with a Service agent connected to Enhanced Chat v2, CLTs in agent actions render as your customers see them.

• Personalize Voice Interactions Faster with Voice Call Variables in Canvas View

Give your voice agents access to call details so that they can respond with more context. After you add the Telephony connection to your agent, you can add voice call variables to it on the Variables page in Canvas view and use the variables throughout your agent’s reasoning and interactions. Previously, you could only use voice call variables with Agent Script.

• Reduce Resolution Time with File Sharing in Enhanced Chat v2

Give service reps the context they need to resolve inquiries faster. Customers can now share files such as screenshots, receipts, and PDFs with service reps in Enhanced Chat v2 sessions. Sharing files with AI agents isn't available yet.

• Create Agents in the New Builder Only

The New Agent button is no longer available in the legacy Agentforce Builder in Setup. You can now create agents in the new Agentforce Builder in Agentforce Studio only. This change doesn’t affect your existing agents in the legacy builder. You can continue to edit, activate, version, and otherwise manage them. We recommend upgrading your agent from the legacy builder to the new builder, which creates a version of your agent in the new builder. The new version includes all of the original agent’s subagents, actions, system messages and settings, data, and connections, converted to Agent Script.

• Agentforce Observability: Create Scorers via API, Interaction-Level Metrics, SOMA and Voice Session Trace Additions, Changes to Billing and Consumption

Custom scorers are now generally available (GA) via API. The Session & Intents view has new columns that showcase Scorers and Error data. Session Tracing includes more granular multi-agent and metrics' breakdown. This release also provides new service level metrics to better gauge your agent's interaction with customers and how they affect your credit usage and billing.

• Guide Your Users to Agent Conversations from Lightning App Home Pages

Welcome recommendations are prompts that suggest ways to start a conversation with an Agentforce Employee agent, so your users can select a prompt instead of typing a message from scratch. Welcome recommendations are available only for Agentforce Employee agents in Lightning Experience, in both Agentforce Builder and the legacy Agentforce Builder. They now appear on supported Lightning app Home pages, including Sales Home, Service Home, Marketing Home, and custom application Home pages. Previously, they appeared only on record pages and list views.

• Control Which Model Each Subagent Uses in Agentforce Builder

Select the model that gives your subagent the best balance of speed and reasoning depth for its specific job. For example, a subagent that performs simple tasks can run on a smaller model and one that manages complex tasks can use a larger one. Previously, you could only set a model for a subagent with Agent Script.

• Deploy Agentforce Employee Agents on Enhanced Chat v2

Give your employees a chat experience optimized for modernized features and agentic use cases. You can now deploy Agentforce Employee agents on Experience Cloud sites with Enhanced Chat v2.

• New Web Search Data Libraries Aren't Supported in the Agentforce Data Library

The Agentforce Data Library no longer supports creating web-based data libraries. You can still view and use your existing web search data libraries.

• Test Prompt Templates in Agentforce Testing Center (Beta)

Surface patterns and edge cases missed in smaller-scale testing when you run tests at scale. Test your prompt templates by selecting Prompt Templates as the test type in Agentforce Testing Center. Build test cases and manage test suites, all in a spreadsheet-inspired interface that makes it easier to validate and debug.

• Toggle Intelligent Context Processing On or Off in Agentforce Data Libraries

Enable or disable Intelligent Context processing for file-based data libraries. Intelligent Context improves retrieval accuracy for tables and infographics, and is enabled by default.

• Get Real-Time Updates on Your Agentforce Data Library Build and Resolve Issues Faster

Track the progress of your Agentforce Data Library build stage by stage, and use links to relevant objects to review and resolve issues faster.

• Choose a Transcription Model for the Speech to Text Agent Action

Use the Transcription Model parameter on the Speech to Text action to control which model converts audio to text. Choose the model that best fits your transcription needs, such as audio quality, language, and data handling requirements. This parameter is available in Agentforce Builder, Flow, and REST.

• Quickly Access Enhanced Chat v2 Deployment Options in Agentforce Builder

Spend less time switching between setup pages for Enhanced Chat v2 deployments. You can now preview an Enhanced Chat channel's deployment or copy its code snippet from the channels table in Agentforce Builder.

• Set Context Variables Faster for More Accurate Agent Previews

When you preview an agent, it’s easier to find where to set initial variables so your agent’s preview behavior accurately reflects your user’s context. Previously, you viewed and edited variable values in the Variables tab in the bottom panel, next to Trace. Now you set and override variable values in the new Context Variables tab, which is more discoverable at the start of a session. We’ve also added an icon to warn you when your agent uses a linked variable that’s set for you in production, but isn’t set in preview—for example, a value that’s filled from the Messaging Session object when your agent is live on a Messaging channel. To track variable values throughout the session, you can still use the Variables tab in the bottom panel.

• Agentforce Observability: Insights Page Retirement and Model Selection

Agentforce Studio's Insights page is retiring in July 2026. Observability data moves to Analytics dashboards for a comprehensive view of session and intent analysis. Agentforce Observability also now uses your org-level LLM provider selection.

• Create AI-Powered Service Channels from Your Knowledge Base

Quickly build AI-powered service channels grounded in your knowledge base using Quick Service Agent Configuration. This process can create your self-service help portal, Voice channel, or web chat and power them with Help agent, built on the Agentforce Service agent template. This process handles setup and prerequisites automatically, letting you launch channels without the overhead of Experience Cloud sites or user profiles. Use these channels to provide 24/7 support and deflect cases.

• Select Gemini for Agentforce

Select Google Gemini for LLM requests in Agentforce. This option is an alternative to the existing Salesforce Default option, which uses OpenAI, and the AWS-Hosted option, which uses Anthropic. The Google Gemini option uses Gemini 3.5 Flash. Google Gemini is supported for agents built in the new Agentforce Builder only. Don't select Google Gemini if you have active agents created in the legacy Agentforce Builder.

• Sharper Transcription and More Natural Voice for Agentforce Voice

Your voice agents now deliver sharper transcription and better natural speech output—no configuration or restart required. Transcription is more accurate in noisy environments, with accented speech, and with specialized terminology. Non-English personas also now support custom International Phonetic Alphabet (IPA) or Carnegie Mellon University (CMU) pronunciations, giving them the same customization options as English-speaking agents.

• Chat with Agentforce in 24 More Languages (Beta)

In addition to English, Japanese, French, German, Italian, Japanese, Portuguese, and Spanish in certain locales, Agentforce now supports 24 more languages.

• Gain Insights Into Agentforce and Generative AI Consumption

Monitor your AI usage and consumption. The new AiAgentGenerativeAiUsage model contains your Agentforce and generative AI usage including metered and unmetered usage. Updates to the data occur every 5 minutes. Write Data 360 queries and build reports and dashboards. Conduct usage analytics and gain actionable insights into AI consumption across your entire organization.

• Edit Subagents in the Asset Library

You can now edit subagents directly in the Asset Library in Setup. Updating a subagent in the Asset Library doesn't impact current agents, which use a copy of the original subagent.

• Agentforce Observability: Refined Agent Analytics and Custom Scorers (Beta)

Get a unified view of agent performance with Refined Agent Analytics, which combines Service Agent Analytics and Employee Agent Analytics in one place with more than 40 metrics. Use Custom Scorers (Beta) to evaluate sessions against your own KPIs alongside Salesforce standard quality metrics.

• Unlock Agent Interoperability with MCP for Agentforce

Empower your agents with capabilities across providers with Model Context Protocol (MCP) for Agentforce. MCP is an open standard for seamlessly connecting AI models to external systems. We’ve built an MCP client into Agentforce, so the complex, manual work of integrating systems–including API connectivity, authentication, and data access–is done for you. MCP for Agentforce and its suite of features makes it easy for you to register and manage MCP servers, create policies to govern agent behavior, and give your agents secure access to MCP server tools.

• Protect Connections to Your External APIs and MCP Servers with Agentforce Policies

Create policies to protect connections to your external API and MCP servers so that agent tools can securely and properly access those services. Control access, set usage limits, and support governance standards.

• Route Agentforce Voice Calls Using SIP

Take advantage of the internet’s cost-effective and flexible virtual connection. Route calls to an Agentforce Service agent using Session Initiation Protocol (SIP). This protocol transmits calls over the internet. Previously, you could route these calls with a Public Switched Telephone Network (PSTN) phone number only.

• Route Agentforce Voice Calls Globally with Dynamic Routing

Expand call routing globally with Dynamic Routing, a new communication protocol for Agentforce Voice. PSTN is limited to the US and Canada. SIP requires a dedicated integration with a vetted telephony partner. Dynamic Routing removes both barriers. It works worldwide and doesn't require a SIP trunk or telecom resale agreement.

• Record Voice Calls for SIP and Dynamic Routing Channels

Record voice calls between customers and the Agentforce Voice agent on your SIP and Dynamic Routing voice channels. Use recordings to support quality assurance reviews, agent coaching, and compliance initiatives. To easily distinguish speakers during playback, recordings are stereo-separated, with the customer on one channel and the agent on the other. Each recording is stored as an MP3 file and automatically linked to its voice call record for easy access.

• Create and Deploy Voice Enabled Agents in the New Agentforce Builder

You can now build, customize, and deploy voice-enabled service agents in the new Agentforce Builder. Use Agent Scripts and multi-step workflows within the new builder to build your voice-enabled agents.

• Expanded Global Language Support with Agentforce Voice (Beta)

Agentforce Voice now supports additional global languages beyond English and French, allowing you to reach a broader audience. These languages are currently supported in Beta.

• OpenAI Search Provider in the Search the Web Agent Action Is Now Generally Available

Get relevant, real-time web search results within your AI agents by using OpenAI as a search provider. OpenAI as a search provider, which is now generally available, includes an update to the underlying web search model since the beta release.

• Increase Trust with Citations in Bring Your Own Channel

Build trust in AI-generated responses by providing source references in Bring Your Own Channel outbound messages. Messaging end users can now verify Agentforce Service Agent's answers against knowledge articles or other sources directly within your custom UI. With source links and character offsets in the outbound message payload, partners can render inline markers or a dedicated sources section.

• Easily Upgrade Agents from the Legacy Builder to the New Builder

Upgrade an agent from the legacy Agentforce Builder in Setup to the new Agentforce Builder in Agentforce Studio, with just a few clicks. Previously, you had to manually recreate your agent in the new builder. Now, you simply specify the agent and version that you want to upgrade, and then we create a new version in the new builder for you. The upgraded agent includes all of the original agent’s subagents, actions, system messages and settings, data, and connections, converted to Agent Script. Plus, after you upgrade, Agentforce can optimize your agent for Agent Script to improve your agent’s reliability and performance.

• Build an Agent Script-Powered Agent End to End with the Updated Implementation Guide

The Agentforce Implementation Guide is our most comprehensive, hands-on resource for building agents on the Salesforce Platform. We’ve updated the guide for the new Agentforce Builder in Agentforce Studio. Follow along to build a more deterministic agent with Agent Script and see the power of hybrid reasoning in action. Take advantage of new testing capabilities with the enhanced preview and debugging experience in Agentforce Builder and the new Agentforce Testing Center in Agentforce Studio. Plus, learn how to get the most out of your agent with an updated Observability guide. Complete the implementation guide to get a fully functional customer-facing agent that can answer customer questions, create and update cases, and escalate to a support rep.

• Transform Your Customer Experience with Voice-Enabled Agents in Enhanced Chat v2

Scale your operations and decrease agent handle time by offering customers a more engaging and accessible way to interact with your service agents. With Agentforce Voice for Enhanced Chat v2, customers can seamlessly switch between text and voice mode. Easily set up a voice-enabled service agent and customize it to meet your unique business needs in Agentforce Builder.

• Elevate Your Enhanced Chat v2 Deployments with More Customization and Reliability

Strengthen your brand presence and keep conversations flowing with Enhanced Chat v2 customization and messaging enhancements. Control the chat window dimensions and background colors. Improve the end-user experience with read and delivery receipts, smoother message rewrites, and improved auto-response messages, including inactivity notifications.

• Build Agents Faster and Ensure Consistency with Connection Inputs for Marketing Email

Reduce setup time and keep experiences on-brand with the new signature and disclosure inputs for the Marketing Email connection. By default, connection inputs and their settings are shared across all agents that have the connection. When you customize a specific agent’s connection inputs, it doesn’t impact other agents that use the connection. Similar to variables, you can use connection inputs throughout your agent’s reasoning and interactions. Input values can’t be changed during a conversation.

• Orchestrate Other Agents (Beta)

Extend your agent’s capabilities with Multi-Agent Orchestration for Agentforce (Beta). Connect your Agentforce agent with other specialized Agentforce agents in your Salesforce org. Together they can collaborate seamlessly on complex tasks. By providing a single, unified point of contact, connected subagents help your users accomplish more without juggling multiple disconnected sessions.

• Prepare for Updated Model for AWS-Hosted Option in Agentforce

The AWS-Hosted option for Agentforce will use Anthropic Claude Haiku 4.5 on Amazon Bedrock later this month. It currently uses Claude Sonnet 4, which is in the Legacy state on Amazon Bedrock.

• New and Changed Agent Script Functionality

Use Agent Script to build agents that blend deterministic logic with the reasoning power of LLMs.

• New and Changed Standard Agent Actions and Subagents

Quickly add powerful functionality to an agent with new and changed standard agent actions and subagents.

Original source

Learn more about our latest release, designed to help you connect to customers, employees, agents, and operations in a whole new way.

Agentforce 360 is our boldest release yet, shaped by thousands of real-world customer deployments.

This release introduces a new Agentforce Builder for faster agent development, Agent Script for greater control of agent behavior, Agentforce Voice for natural on-brand conversations, and Intelligent Context to ground agents in your complex unstructured data.

Explore Agentforce 360

Agentforce Builder

Agent Script

Agentforce Voice

Intelligent Context

Everyone can now build and control reliable agents with Agentforce Builder

With Agentforce Builder, teams can now craft, test, and refine agents in a single, conversational workspace. Whether you're a seasoned developer or a natural language builder, you can ditch the separate build, test, and deploy cycles. Revolutionize your workflow with "vibe-building," instant editing in doc-like, canvas, or script views, and one-click simulations with real-time debugging. Iterate radically faster, safer, and more collaboratively than ever before. Every agent compiles into a portable JSON file, making versioning and sharing a breeze, while our enterprise-grade security and privacy controls ensure top-tier governance.

Key capabilities include:

• Unified workspace: Collapse drafting, testing, and deployment into a single interface, eliminating the traditional build-test loop.
• Multi-view editing: Switch seamlessly between a doc-like editor with autocomplete in a low-code canvas, and a script view to match any workflow.
• Rapid iteration with simulation and live debugging. Preview real user interactions in real time, view trace data for each reasoning step, and adjust instantly.
• Enterprise-grade governance: Built on Salesforce’s security and privacy features to ensure compliance and safe deployment.
• Agentforce Vibes: An AI coding partner that collaborates like a pair programmer, understanding your Salesforce project context and executing on your behalf to accelerate development.

Control agent behavior with Agent Script

Agent Script is a new scripting language for controlling agents. It combines the creativity of AI with the predictability of code so that agents perform the exact way you want them to, every time. Agent Script allows you to define agent behavior with a human-readable expression language, which enables conditional logic, precise tool use, and guided, deterministic controls. This new level of control empowers Agentforce to tackle even your most complex tasks with confidence.

Key Capabilities include:

• More predictable agents - Agent Script is a new scripting language designed for controlling agents. This human-readable, portable JSON expression language allows teams to define complex agent logic, including conditionals, “if/then” rules, and hand-offs - all with programmatic precision.
• Hybrid Reasoning in the Atlas Reasoning Engine: The Atlas Reasoning Engine is now configurable, enabling teams to balance the inherent creativity of LLMs with the certainty of structured business logic, enabling more predictable and reliable agent behavior.
• Flexible, expanded model choice: Agentforce now supports Google Gemini as a model for the Atlas Reasoning Engine, joining Open AI and Anthropic on Amazon Bedrock to enable even greater choice for what models Agentforce uses to reason and plan.

Create unified, dynamic experiences that delight with Agentforce Voice

Leave long waits and frustrating phone menus behind. Agentforce Voice delivers natural, human-like conversations that capture your brand's unique tone and personality, and connect seamlessly to every part of your business. With 80% of all inquiries starting by phone, we're transforming every call into a dynamic, real-time interaction that is fully integrated into every Salesforce application. We put each agent through thousands of real-world scenarios—including noisy backgrounds, distracted callers, and complex, multi-step processes—to ensure reliable, enterprise-ready performance. The result? Stronger customer relationships and measurable ROI.

Key capabilities include:

• Fully personalized voice agents: Each AI agent speaks in a voice aligned to the company’s brand, trained on its data, tone, and contextual knowledge.
• Ultra-realistic, low-latency interactions: Built for real-time responsiveness, Agentforce achieves near-human conversational flow while executing actions such as updating CRM records, triggering workflows, or calling APIs - with full auditability through Data 360.
• Seamless integration: Natively connected to Salesforce, Agentforce Voice extends across the contact center stack with first-class support for Amazon Connect, Five9, Genesys, NiCE, and Vonage.
• Fluid, interruptible conversations: Every word is transcribed live in Salesforce Voice, giving human agents real-time visibility from within the agent console to allow instant, seamless takeovers when needed.
• Voice-first platform architecture: Enhanced latency, speech realism, and deep language understanding make it not just a new modality - but a complete platform experience built to meet customers where they are.

Make every agent more accurate with Intelligent Context

Enterprise-grade agent experiences start with business-specific context. Intelligent Context ensures your AI agents are deeply grounded in the complex, unstructured data of your business. Powered by the new Data 360, this feature automatically extracts, structures, and surfaces this information so your agents can deliver accurate, relevant, and business-specific answers. Our low-code pipeline lets you configure document processing in hours instead of days. Plus, you can configure context so agents interpret the same content from multiple business perspectives. The result is agents that reason with the right data, empowering your teams with rich, contextual intelligence in every interaction.

Key capabilities include:

• Automated context extraction: Converts complex, unstructured documents into structured, actionable data for Agentforce.
• Low-code configuration: Quickly set up unstructured data pipelines in hours, not days, without extensive engineering effort.
• Multimodal support: Handles PDFs, tables, images, flowcharts, and other diverse content types.
• Seamless integration with Data 360: Create a search index with AI, then activate it everywhere for secure, contextual, enterprise-ready operations.

Build, deploy and manage AI agents at scale.

Learn more: Agentforce Builder

Configure Agentforce Topics, Subagents, and Instructions in a few clicks.

Learn more: Agentforce Voice

Bring AI-powered voice to every channel with Agentforce Voice. Deliver fast, natural conversations across phone, web, and mobile.

Learn more: Multi-Agent Orchestration

Build a collaborative AI agent team that works together to solve complex problems faster.

Learn more: Agentforce Observability

Monitor, analyze, and optimize agent performance in near real time so you can deliver faster resolutions, higher accuracy, and better customer experiences.

Agentforce 360 Ecosystem

Customer Model Choice for Agentforce

Agentforce Voice CCaaS Partners

Ready to take the next step with Agentforce?

Build agents fast.

Take a closer look at how agent building works in our library.

Get expert guidance.

Launch Agentforce with speed, confidence, and ROI you can measure.

Talk to a rep.

Tell us about your business needs, and we’ll help you find answers.

Original source