Biggest Dreamforce ’25 Announcements: Everything in a Nutshell
Dreamforce 2025 was a celebration of everything “agentic” – Salesforce’s vision for a world where humans and AI agents work side by side. The message was loud and clear: this isn’t just another AI wave, it’s the next evolution of how businesses operate.
In this article, we’ll unpack and demystify the biggest themes, announcements, and insights to come out of Dreamforce ‘25.
The Agentic Enterprise Comes of Age
Marc Benioff opened Dreamforce by declaring that the “Agentic Enterprise” is here to stay. This means companies will no longer see AI as an add-on but as an active colleague. The focus was on trust, data, and governance, the foundations of safe, reliable automation. In short, AI should help, not hinder, and do so transparently.
Meet Agentforce 360
The headline act of Dreamforce ‘25 was Agentforce 360, the new centerpiece of Salesforce’s AI strategy. Think of it as the connective tissue across Sales, Service, Marketing, Commerce, Slack, and Tableau. Every app can now tap into the same AI models and data, ensuring consistent reasoning and reliable results.
Agentforce 360 combines four key layers:
- Agentforce Platform: Where agents are built, tested, and deployed – now with voice interaction and a conversational “Builder” workspace.
- Data 360 (formerly Data Cloud): The data intelligence layer that gives agents context and governance.
- Customer 360 Apps: Salesforce’s familiar applications, now infused with AI that understands business processes deeply.
- Slack: The conversational front door where humans and agents collaborate in real time.
Data 360
Data 360 (formerly Data Cloud) got a full reintroduction, positioning it as the “brains” of the platform. It turns scattered, unstructured data such as PDFs, tables, and images into useful context for AI agents.
Its Intelligent Context feature lets agents understand and reason with the same data a human would, while Tableau Semantics ensures everyone in the business is using the same language for metrics.
Slack
Slack continues to evolve as the “Agentic OS” (Agentifc Operating System). It’s now the main workspace for humans and agents alike. A completely reimagined Slackbot acts as a personal AI companion, summarizing threads, drafting messages, and surfacing Salesforce insights.
Through the new Model Context Protocol (MCP), Slack can now connect with third-party AI tools like Anthropic and OpenAI, bringing all your favorite assistants together in one workspace. MCP is what enables multi-agent interoperability. In simple terms, it lets different AI agents talk to each other, share information, and trigger actions across multiple systems.
Acting as a universal connector, MCP gives your Agentforce agents the freedom to pull in data or perform tasks beyond Salesforce, safely and consistently. In essence, it’s the bridge that links Salesforce to the wider world of AI.
OpenAI Partnership
In one of the biggest surprises of the event, Salesforce and OpenAI announced a deep partnership. GPT-5 is now available directly inside the Salesforce platform, while Agentforce 360 data and insights can surface inside ChatGPT. This means users can ask questions like “How are our Q3 sales tracking?” inside ChatGPT and get real answers from Salesforce data, securely and with full context.
The partnership also powers new experiences such as Agentforce Commerce in ChatGPT, where customers can browse and buy directly within a chat. It’s a huge leap in blending consumer and enterprise AI.
Agentforce Voice
One of the most exciting reveals at Dreamforce ’25 was Agentforce Voice, Salesforce’s leap into natural, conversational AI (technically, I would argue this is their second leap…remember Einstein Voice Assistant?). While chat-based agents have been around for a while, Agentforce Voice changes the dynamic entirely, making interactions sound and feel more like real human conversations.
In simple terms, Agentforce Voice lets businesses build voice-based agents that can answer calls, resolve issues, and even carry out transactions, all while staying connected to Salesforce data in real time. The result is a far more natural experience for customers and service teams alike.
Behind the scenes, Agentforce Voice combines low-latency speech synthesis, live transcription, and deep integration with Salesforce records. That means when a customer calls about an order or a service issue, the AI can instantly pull their data, understand their intent, and take action, whether that’s updating a record, triggering a workflow, or handing over to a human agent when needed.
Agentforce Voice supports multiple integrations out of the box, including Amazon Connect, Five9, NiCE, and Vonage, making it easier to embed directly into existing contact center systems.
Agentforce Builder
Salesforce understands that building can sometimes feel like juggling too many tools at once: flows here, setup there, prompts somewhere else. That’s exactly what Agentforce Builder is designed to simplify. With its new conversational interface, design, testing, and deployment now happen seamlessly in a single workspace.
To be clear, Agentforce Builder isn’t new – what’s changing is the user experience. From now on, the way you build and test an agent will be more intuitive, streamlined, and connected than ever before.
Think of it as a command center for your AI agents. You can describe what you want the agent to do in natural language, for example, “Create an onboarding assistant that checks new-hire tasks and sends reminders,” and the Builder translates that into a working agent.
The interface offers three flexible views: a doc-like editor for natural-language drafting, a low-code canvas for visual building, and a script view for developers who want fine-grained control. You can switch between them instantly, meaning admins, developers, and architects can all collaborate without losing context.
What makes the Builder especially powerful is its real-time simulation. You can test your agent’s reasoning, watch each decision step unfold, and adjust instantly before going live. Salesforce calls this “vibe-building”, rapid iteration powered by conversational design.
Setup Powered by Agentforce
Admins have long said that Salesforce Setup takes too many clicks, and it’s tough to navigate an area that has 1,200+ pages! The new Setup Powered by Agentforce lets admins use natural language to manage users, create fields, fix permissions, and even build flows.
With the Winter ’26 release, Salesforce gave Setup Home a major refresh, introducing clearer navigation and a cleaner layout. Agentforce takes that even further with a collapsible sidebar for easier access to tools, live org health, and usage metrics right on the homepage, and a new prompt bar where you can chat directly with the Setup agent to get work done faster.
You can ask, “Does Emily White have access to Accounts?” and the agent will check and act accordingly. This is one of those quietly revolutionary changes that will save admins hours each week.
Agentforce Vibes
Dreamforce wasn’t just for admins. Developers saw the rise of Agentforce Vibes, a conversational AI coding assistant that can generate Lightning Web Components or Apex logic from natural language.
What makes Vibes different is its contextual intelligence. Because it’s integrated with your existing Salesforce metadata, Vibes doesn’t just generate code; it builds solutions that make sense within your org. It can reuse components, follow naming conventions, and check dependencies before deploying.
For developers, Vibes functions like a “pair programmer.” It can generate Lightning Web Components, write Apex logic, or refactor code to follow Salesforce best practices. Developers can review the output in real time, switch to a script view for fine-tuning, or test their work instantly through one-click simulation.
Each org gets 50 daily calls to the premium model (currently GPT-5), before falling back to an open model, so most teams can experiment freely. It’s also available as an extension for VS Code and inside Code Builder, making it easy to plug into existing workflows.
You can absolutely experiment with Agentforce Vibes even if you’re not a developer, but it’s worth doing so with a bit of caution. During Dreamforce, Tim Combridge and I tested the Vibes hands-on experience and found it to be a little temperamental. It often ran slowly and occasionally got stuck trying to correct its own mistakes. Since the event, Tim has continued testing Vibes at home while attempting to build a Flow, and he’s still run into similar reliability issues. (Stay tuned for his full write-up on that soon!)
Agent Script
If Vibes is about creation, Agent Script is about control. It’s Salesforce’s new scripting language for defining how AI agents think, act, and respond.
Agents, by nature, are flexible and adaptive. That’s part of their power, but it also introduces risk. You don’t always want an AI improvising when handling compliance workflows or sensitive customer data. Agent Script solves that problem by letting you inject conditional logic, like those familiar “if/then” rules, directly into an agent’s reasoning.
Written in human-readable JSON, Agent Script allows teams to combine deterministic workflows (where every outcome is predefined) with the creative reasoning of large language models. For example, you might define that an agent must always check credit limits before approving a quote, but it can phrase its explanations however it chooses.
The result is hybrid reasoning: AI that’s structured where it matters and flexible where it counts. You can set boundaries, specify tool use, and hand off between agents safely, all while maintaining transparency.
Salesforce’s Atlas Reasoning Engine is now configurable to work with Agent Script, meaning you can choose how much “freedom” to give your agents and even select which language model they use. Every decision and reasoning step can also be observed through the new Observability Dashboard, so you can see exactly why an agent did what it did.
Agentforce for Sales
For sales teams, Salesforce’s message this year was simple: AI isn’t just automating data entry, it’s becoming a sales partner. Agentforce for Sales brings the concept of the “Agentic Enterprise” directly into the sales cycle, from prospecting to renewal.
Everywhere across the experience, Agentforce acts like a virtual co-seller. It can qualify leads, suggest next-best actions, and even build quotes automatically. Because it runs on top of Data 360, every recommendation is grounded in live, governed data, meaning your agent knows not just who to call, but why.
Sales reps can ask natural-language questions inside Salesforce or Slack, such as “Show me high-value opportunities at risk this quarter” or “Draft a follow-up email for yesterday’s demo.” The agent then surfaces relevant context, generates the content, and logs the activity. It’s all conversational, transparent, and fully auditable.
The biggest change, though, is that the sales experience is now continuous. Agents maintain context across channels, from a website chatbot to an email thread to a phone conversation via Agentforce Voice. That means no more lost history or repeated introductions.
For sales reps, the new Sales Workspace adds a command center-like experience for individual sellers to transform the way they work. Alongside the embedded Tableau dashboard and Opportunities recommendations at the top, Agentforce also generated a list of actions where human input is needed. On top of that, users will be able to see, at a glance, all tasks completed by agents across all key areas within sales cycles.
For managers, the new Agentforce Command Center visualizes performance, pipeline health, and conversion trends in real time. It can even alert leaders when deals stall or sentiment shifts. Instead of chasing data, teams can act on it instantly.
Agentforce For Service
Service Cloud has been reborn as Agentforce Service, a proactive service platform built around AI assistance. At the heart of the update is the new Command Center for Service, a real-time control hub where leaders can see every customer interaction across chat, email, voice, and social. It tracks sentiment, volume, and trends, surfacing insights before they turn into issues.
For frontline staff, the familiar Service Console has evolved into an intelligent workspace. It automatically summarizes customer history, highlights key details, and uses a new Service Rep Assistant to recommend next steps or verify data on the fly. If an agent encounters an unfamiliar problem, the assistant searches past cases to find similar resolutions and can even create new knowledge articles afterward.
Customers, meanwhile, benefit from the Agentforce Service Portal, a conversational self-service experience that feels genuinely human. It can handle bookings, troubleshoot issues, or make proactive recommendations, all powered by Data 360 for personalized, context-rich responses.
Summary
Dreamforce ‘25 showed us that AI at Salesforce isn’t a bolt-on; it’s the platform itself. From admin setup to developer tools and customer experiences, Agentforce is woven throughout.
If Dreamforce ‘24 was the debut of Agentforce, Dreamforce ‘25 proved it can scale: contextually, and across every corner of the business.
In Benioff’s words:
“We’re entering the age of the Agentic Enterprise — where AI elevates human potential like never before.” Marc Benioff, CEO Salesforce
And judging by what we saw this year, that future has already begun.
