What actually shipped at Dreamforce '25 — Agentforce Builder, Agent Script, hybrid reasoning and Voice — and where a developer should start.
On October 13, 2025 at Dreamforce, Salesforce announced general availability of Agentforce 360 — a rebrand and major upgrade of the Agentforce platform positioned as "the agentic enterprise" layer: one trusted system connecting humans, AI agents, apps and data. In practice, the umbrella covers four things developers care about: the Agentforce 360 Platform (agents + builder + reasoning), Data 360 (the evolution of Data Cloud), Customer 360 apps with agents embedded, and Slack as a first-class conversational surface.
It is not a separate product you buy from scratch — existing Agentforce orgs move onto the 360 naming and gain the new capabilities as they roll out through the Winter and Spring '26 release cycles.
Agentforce Builder is the new development studio that replaces the classic Agent Builder experience. You describe the agent you want in natural language, and Builder scaffolds topics, instructions and actions — then lets you test and refine in the same canvas. Under the hood it produces the same agent metadata you can version-control and deploy with standard ALM tooling.
The most important developer feature in the release: the Atlas Reasoning Engine now supports hybrid reasoning, and you program it with Agent Script — a declarative, human-readable DSL. The idea: LLMs are great at understanding people but enterprises need strict, sequential, predictable steps for things like refunds, eligibility checks and compliance flows. Agent Script lets you pin down the deterministic parts while leaving conversation handling to the model.
The core constructs:
@action.Your_Action_Name bindings and typed results.@variable.user_city) inject live context into reasoning instructions.@utils.transition to) hand off between topics — the building block of multi-agent flows.topic weather_forecast:
description: "Answer questions about the weather for a city"
reasoning:
instruction: |
The user lives in {{@variable.user_city}}.
Fetch the forecast before answering:
{{@action.Get_Weather_Forecast}}
actions:
Get_Weather_Forecast:
with city = @variable.user_city
set @variable.forecast_data = @result.daily_forecast
if @variable.forecast_data is empty:
@utils.transition to @topic.human_handoff
Agentforce Voice is a native voice layer for agents: low-latency transcription, realistic speech synthesis and direct access to your Salesforce data and actions. The pitch is replacing menu-tree IVR ("press 2 for billing") with an agent that resolves the call. If you have built Service Cloud Voice integrations before, note this is agent-first — the same topics and actions you build for chat power the phone channel.
Real-life scenario: a telecom support line replaces its IVR with a Voice agent. The agent authenticates the caller against CRM data, answers "where is my technician?" from Field Service records via an action, and only transfers to a human when the caller asks for something outside its topics — with a full transcript attached to the case.
Data 360 is the new name and expanded scope for Data Cloud — the grounding layer for agents: zero-copy access to warehouse data, unified profiles, and retrieval for RAG-style answers. Meanwhile Slack becomes a primary agent surface: Agentforce agents (and the new Slackbot experience) run where teams already work, so an employee-facing agent can be rolled out with no new UI at all.
Sources: Salesforce Agentforce 360 announcement (Oct 13, 2025) · Introducing Hybrid Reasoning with Agent Script · Agentforce 360 announcements hub