Salesforce Announces Agentforce 3: Command Center, MCP, and Apps

Salesforce has just announced Agentforce 3.0 – the latest iteration of its flagship AI product.

Agentforce has been Salesforce’s biggest talking point since it first launched in September 2024, quickly becoming a central pillar of their AI strategy. We’ve seen a steady stream of updates since then, with Agentforce 2.0 arriving in December, followed by Agentforce 2dx in March. Now, the Agentforce train keeps on rolling with the release of Agentforce 3.0, another major milestone.

This latest update signals that Salesforce is listening closely to the ecosystem. Many of the concerns raised by professionals over the past few months around visibility, control, and integration are being directly addressed. 

Agentforce 3.0 introduces features that give users deeper insight into agent performance, greater governance over their deployment, and more seamless integration with the tools and workflows businesses already rely on.

Let’s take a closer look at what’s new in Agentforce 3.0 and what it means for Salesforce professionals looking to get started with AI agents.

Why Agentforce 3.0 Matters

When Agentforce 2.0 was introduced late last year, it felt as though we were seeing the first steps of Salesforce’s vision of a digital labor workforce come to life. The company introduced new pre-built agent skills for Slack, Tableau, and AppExchange partners, Agentforce in Slack, MuleSoft integrations, and an upgraded Atlas Reasoning Engine, amongst many other features.

But from our perspective, the sentiment from Salesforce professionals around Agentforce has remained skeptical – with concerns surrounding security, technical debt, pricing, governance, and ROI.

However, the move from the $2-per-conversation model to Flex Credits, combined with the newly announced general availability of dedicated Agentforce user licenses and add-ons, shows just how committed Salesforce is to driving adoption and making it easier for customers to get started. It’s a clear signal that they’re aiming to remove friction and scale usage across teams.

And now, Agentforce 3.0 takes this a step further. With new ways to monitor your agents, as well as the opportunity to connect to external enterprise tools, Agentforce is starting to move away from just being a cool prototype and becoming an enterprise-grade AI tool that it’s been hyped up to be.

Agentforce Command Center: Monitor Your Agents

One of the key features of this new update is the Agentforce Command Center, a new way to track, monitor, and measure your AI agents.

The Command Center allows you to:

• Monitor agent health in real time: Track performance, error rates, and escalations with alerts that flag issues instantly.
• Understand adoption at a glance: See what agents are used most, what’s working, and where improvements are needed.
• Spot patterns and optimize fast: Analyze interactions and get AI-driven recommendations to improve agent performance.
• Trace every interaction: Built-in session tracing (via OpenTelemetry and Data Cloud) enables enterprise-wide observability.
• Tailor Command Centers for every team: View AI and human activity side-by-side, starting with Service Cloud wallboards.
• Build and test faster: Use natural language to generate agents, and simulate performance at scale with built-in testing tools.

As humans begin working more closely with AI agents, the Command Center introduces a critical observability layer which allows users to monitor agent health, track performance, and continuously refine agents at scale.

ROI has been a sticking point for many. As Ian Gotts, CEO of Elements.Cloud, previously put it: “Nobody wants to hand out a blank cheque – they want caps and predictable ROI.”

Until now, the lack of visibility into how agents were performing has held back adoption for some since Agentforce first launched. But with the Command Center, users finally have a clear view of what their agents are doing, making it easier to measure impact, optimize performance, and understand exactly what they’re getting in return.

The Command Center will become generally available this August.

Model Context Protocol: The USB–C for AI Agents

AI agents are only really useful if they can actually do things inside the tools your business uses, like send files, update records, or trigger workflows. But safely plugging them into all of those systems without breaking security rules or writing lots of code might leave many in a tricky situation. 

Salesforce is looking to solve this with the Model Context Protocol (MCP), which the company describes as a universal plug, “like USB-C for AI agents.”

This new feature will allow agents to connect to different systems without the need for custom setups, ultimately making integrations quick and secure.

The MCP will also enable you to:

• Instantly convert APIs into agent-ready assets with MuleSoft: Using new MCP connectors, MuleSoft can transform any API or integration into an MCP-compatible service, complete with security, monitoring, and traffic control, making it easy to build and manage multi-agent workflows across your stack.
• Host custom MCP servers with Heroku AppLink: Developers can spin up and expose custom MCP services using Heroku, offering a fast, secure, and scalable way to connect bespoke tools to Agentforce using Heroku’s infrastructure and automation.
• Enable secure agent actions inside Slack: Slack will soon offer its own MCP server, built with Anthropic, letting Agentforce agents interact with Slack messages and files to deliver context-rich insights and take action, all while fully respecting enterprise permissions and data boundaries.
• Control it all with a unified agent gateway: Agentforce introduces a governed gateway, engineered by MuleSoft, that centralizes agent registration, identity, and policy management. Admins can tightly control which agents connect to which tools, how they behave, and ensure all activity follows enterprise-grade security and compliance standards.

This feature in particular shows real signs of how Agentforce is looking to scale, moving beyond isolated use cases into a fully integrated, enterprise-ready platform. Agents should now be able to connect to more tools and take real action, all while playing by your company’s security rules.

The Expansion of the AgentExchange

Acting as an AI alternative to the AppExchange, the AgentExchange was introduced in March to help companies rapidly create and deploy AI agents to improve productivity, efficiency, and innovation across all industries.

The concept fit the Salesforce ecosystem perfectly, building on the existing popularity of the AppExchange and reimagining it for the AI era. Platforms like this encourage collaboration, which has always been one of the ecosystem’s biggest strengths.

With this recent update, it feels as though the AgentExchange is really starting to take shape and could start helping customers with real high-value use cases from well-known companies. This recent update will leverage the previously mentioned MCP servers through the AgentExchange through different partners.

Salesforce provided some serious real-life use-cases, such as:

• PayPal: Service teams can now handle invoice generation via PayPal’s MCP server and agentic commerce capabilities. This will enable Agentforce to process transactions, create compliant invoices, and notify customers over email with the attached invoices, all without manual input.
• WRITER: By securely connecting to the WRITER MCP server, Agentforce will allow WRITER’s AI agents to automate content, retrieve knowledge, and provide compliance checks all within Salesforce workflows.
• Box: Agentforce will support content teams by automatically retrieving the latest signed NDA from Box, summarizing the key terms, and sending a Slack update to the account team with the document and recommended next steps.

While this may be just a handful of use cases, it’s exciting to see Agentforce tied with big-name companies. It’s a big step forward. AgentExchange has gone from being an early-stage directory to a proper, scalable ecosystem. 

Atlas Reasoning Engine Receives Further Enhancements

The Atlas Reasoning Engine acts as the brain behind Agentforce, generating a plan based on what the role is trying to do, then evaluates and refines its plan.

Salesforce initially updated Atlas in their 2.0 update, with enhanced reasoning powered by enriched data from Data Cloud that allowed it to handle more nuanced queries, loop through tools and data sources for better answers, and use retrieval-augmented generation (RAG) with added metadata for improved accuracy.

The 3.0 update takes this a step further with an “enhanced Atlas architecture”. Salesforce says this will make the Agentforce experience: 

• Faster and more responsive: Agentforce now runs up to 50% faster than it did in January 2025, with real-time response streaming so users can see answers as they’re generated.
• Smarter and more accurate: Agents can now use web search to find more relevant information and show inline citations for added transparency.
• Wider model choice: In addition to OpenAI, you can now use Anthropic’s Claude models, hosted securely on Salesforce infrastructure via Amazon Bedrock, which is ideal for highly regulated industries. Later this year, Google’s Gemini will also be supported.
• More global and multilingual: Agentforce is now available in Canada, the UK, India, Japan, and Brazil, and supports six new languages, with over 30 more on the way.
• Always-on reliability: If one model slows down or fails, Agentforce will automatically switch to another in the background, ensuring continuous performance.
• Now approved for government use: Agentforce has achieved FedRAMP High authorization, making it trusted and available for U.S. public sector organizations through Government Cloud Plus.

The improvements to Atlas act as the “icing on the cake” to the update, Salesforce says, bringing faster performance, smarter responses, and greater reliability to everything built with Agentforce.

Other Information

Agentforce 3.0 introduces several updates intended to help teams get started quickly and realize value faster, no matter their industry:

• 100+ pre-built industry actions rolling out in Summer 2025, covering use cases like patient scheduling, advertising proposal generation, and vehicle servicing.
• Out-of-the-box readiness means teams can deploy AI agents with minimal setup across a wide range of sectors.
• Simplified, flexible pricing with new SKUs for Sales, Service, and Industry Cloud
• Per-user pricing with unlimited usage for employee-facing agents, making it easier to scale without worrying about usage limits.

Final Thoughts

As Salesforce continues its goal of scaling Agentforce as quickly as possible, these new features should really help customers optimize their agents and make the most of their potential ability. This update in particular does feel like Salesforce is bringing things together, taking in constructive feedback and acting on it with impact.

However, it will be interesting to see whether Salesforce continues to scale at such a rapid pace. Many have raised concerns about how quickly the “Agentforce train” has been moving. While this update brings plenty of positives, it may also be the right moment to pause, let things settle, and give users the space to fully implement and make the most of what Agentforce offers today.