Agentforce Keynote at Dreamforce 2025: AI Gets Smarter

The Dreamforce 2025 conference brought a huge change to how companies use artificial intelligence. For a long time, the biggest question has been: can we trustAI to handle sensitive customer data and important workflows?

The Agentforce Keynote finally gave a clear “Yes” to that question.

Salesforce spent this year focusing on how to take AI agents—those smart, automatic assistants—out of the testing phase and put them to work safely in big companies. This is a massive step forward from last year.

In this simple review of the Agentforce Keynote at Dreamforce 2025, we break down the three main reasons why Agentforce is now reliable enough for your most important business tasks: Control, Transparency, and Fast Action.

The main issue with AI is that it can sometimes be unpredictable—it likes to improvise and adapt. While adapting is good, it can be scary when an AI agent is handling a mission-critical process. The Agentforce Keynote showed off new tools to solve this problem right away.

Agent Script: The Rulebook for AI

The biggest announcement was the release of Agent Script. You can think of this as the non-negotiable rulebook you write for your AI employee. It lets you add fixed, clear rules directly into the agent’s logic.

For example, if you have an agent that handles expense reports, you can tell it: “IF the amount is over $500, THEN send it to the manager for a quick look, OTHERWISE process the payment right away.”

This new system mixes the creativity of AI (like writing a custom email) with the certainty of fixed rules (like checking security permissions). This is what makes Agentforce safe for finance, HR, and other teams that need strict process control.

Conditional Controls

Salesforce also introduced new conditional controls. These let you use simple IF/THEN logic to guide the agent’s steps. By making the AI’s behavior more predictable, Salesforce is making a strong case that you can rely on the platform for your most core business needs.

To trust AI, you have to see what it’s doing. Before now, how AI made a decision was often hidden, like a “black box.” This made it impossible to fix problems or prove compliance. The Agentforce Keynote focused heavily on opening up this box.

Cleaner Builder and Visual Simulator

The tool used to build agents, the Agentforce Builder, got a major update. It now features a cleaner, simpler screen and a new visual simulator that looks just like a real customer conversation.

This change is essential because it allows builders to test agents quickly and easily. Instead of having to run large, time-consuming tests, teams can simulate the interaction right within the builder. This saves time and helps make sure the agent is perfect before it deals with a real customer or critical record.

The third main point from the Agentforce Keynote is that agents can now act immediately and work better across all your business systems.

The goal is to provide high-quality voice interactions that can smoothly pass a call and all its context to a human expert whenever the situation gets too complex for the AI alone.

Agentforce Grid for Group Management

The new Agentforce Grid is like a super-spreadsheet for managing the work your agents create. If you have lots of records or edge cases that need human review, the Grid lets you review, approve, or adjust them in bulk. This is a huge time-saver for operations teams dealing with high volumes of data.

The enhancements at Dreamforce 2025 transform Agentforce into a powerful tool that focuses on being safe and predictable, making the Agentic Enterprise a real, practical solution.

The Agentforce Keynote at Dreamforce 2025 made it clear that agentic AI is ready for prime time. By focusing on Control (Agent Script), Transparency (Observability), and Action (Voice GA), Salesforce has given companies the confidence they need to use these powerful new assistants in their most important business functions.

This means companies can move past small tests and start building automated teams that are trusted, predictable, and highly efficient. The future is all about helping human talent by giving them smart, safe, and collaborative AI teammates.