Complete Roundup of the Agentforce Keynote at Dreamforce ’25

As far as Dreamforce keynotes go, the Agentforce one was one of the most anticipated, along with the main keynote and the Data 360 one – and yes, Data Cloud got renamed, again. Dreamforce ‘25 showcased Agentforce as a maturing platform, moving from flashy demos toward enterprise-grade capabilities. This year’s keynote focused on powerful enhancements to existing Agentforce capabilities rather than a complete overhaul – and that’s probably for the better. But while the keynote delivered some impressive updates, there are still questions about scalability, governance, and real-world adoption.

In true Dreamforce fashion, the keynote opened with the usual pizzazz about Dreamfest and the Trailblazer community, followed by the presentation of a golden hoodie to Joey Monroe, a consultant from BlueGator. After a hype video showcasing Agentforce success stories, Agentforce EVP Adam Evans took the stage with the keynote titled “Agentforce – Become an Agentic Enterprise.” Unsurprisingly, all the updates focused on giving Agentforce enterprise-grade controls and capabilities. Let’s explore these in more detail.

Watch the Dreamforce ‘25 Agentforce keynote on demand on Salesforce+.

Promise of Enterprise-Ready Agentic AI

One of the key challenges with Agentforce, and agentic AI in general, has been the lack of control over outcomes. Unlike traditional rule-based automations, agents are non-deterministic by nature. They adapt and improvise – they don’t always stick to the script. And that is a problem for large enterprises. Can you trust an AI agent to handle a business-critical process, such as customer verification or purchase reimbursement, correctly every time?

The keynote positioned Agentforce as more controllable and predictable, which is critical for enterprise adoption. Two new concepts stand out: conditional controls and context engineering. Users can now inject IF THEN statements directly into an agent’s reasoning to guide topic and action sequences. Agent Script adds another layer, letting developers fine-tune the balance between consistency and adaptability through their preferred integrated development environment (IDE). These are meaningful steps toward reducing the unpredictability that has made some organizations hesitant.

Clean, Updated User Experience

Salesforce has given the Agentforce Builder a much-needed refresh. The interface is cleaner, and there’s more transparency into what the Atlas Reasoning Engine is doing behind the scenes. This matters because one of the biggest criticisms of agentic AI has been its “black box” nature. Transparency means teams can understand why an agent made a decision, which is essential for trust and troubleshooting. 

Moreover, the new builder provides a visual simulator and actionable recommendations for agent optimization. Previous iterations included a simple checklist of potential issues, but the new version is more immersive. It has the look and feel of a real interaction instead of a basic preview. This allows you to do more of the agent testing within the builder itself, saving time and flex credits you would otherwise spend on batch tests.

Agentforce Voice Is Now GA

Agentforce Voice is now generally available, bringing voice interaction into the platform in a more polished form. You can configure tone and style directly in the Builder, and transcription accuracy is solid. The demo showed that agents can even handle interruptions, which makes conversations feel more natural. Voice is also fully configurable with adjustable settings for tone, accent, use case, gender, and more.

That said, latency is still noticeable. For low-volume or internal use cases, this may not be a deal-breaker, but real-time, high-volume scenarios will need more testing. Voice is a promising addition, but it is not yet clear how it performs under enterprise-scale conditions. Access to Agentforce Voice for partners is currently limited, but testing and developing voice-enabled agents will likely become easier in the near future.

Introducing Agentforce Grid and Observability

Agentforce Grid introduces a spreadsheet-style interface for managing workflows and agents. It allows teams to evaluate and act on multiple cases in bulk, which could be useful for edge cases or fallback processes. However, the demo only showed small-scale scenarios, so it is unclear how well Grid handles thousands of records or how it complements existing tools like Flow.

Agentforce Observability aims to give teams end-to-end visibility into agent performance and business impact. This includes session-level insights, called Moments, and agent-level ROI tracking, which are critical for enterprise adoption. The concept is promising, but the question is whether these insights will be actionable enough to improve reliability at scale. These features are currently in Beta testing, but will become generally available in November.

Other Key Features and Updates

Beyond the headline announcements, Salesforce introduced several smaller but important enhancements that deserve attention:

• Agent Interoperability: Native support for MCP, A2A, and Agentforce sub-agents improves collaboration between agents and external systems. Previously, handoffs between several Agentforce agents have been clumsy and prone to error, so it is good to see this being addressed.
• Improved Multimodality: Enhanced unstructured data handling through Data 360 allows agents to interpret richer content, such as product brochures with text and images.
• Agentforce Vibes: Salesforce introduced “enterprise-ready” vibe coding to let teams build and refine agent behavior using natural language. While Salesforce claims strong guardrails and quality checks, security concerns remain, and their effectiveness in practice is still uncertain. There was an extended demo of Agentforce Vibes in the Developer Keynote.

Final Thoughts

Salesforce has clearly gone above and beyond to position Agentforce as enterprise-ready. The new controls, observability tools, and workflow enhancements show a serious commitment to making agentic AI viable for mission-critical processes. Time will tell if these features deliver on that promise outside the keynote stage. For now, the future of Agentforce looks promising.

However, the emphasis on large enterprise capabilities raises a new question – is Salesforce still a viable choice for small and mid-sized businesses? As the platform becomes more complex and enterprise-focused, SMBs may find themselves questioning whether Agentforce – and the broader Salesforce ecosystem – still fits their needs.