Salesforce Launches Agentforce 3: Enhancing AI Agent Visibility and Interoperability for Enterprises

Salesforce launches Agentforce 3 with AI agent observability and MCP support

Join the event trusted by enterprise leaders for nearly two decades. VB Transform brings together the people building real enterprise AI strategy. Learn more Salesforce rolled out sweeping enhancements to its AI agent platform Monday, addressing the biggest hurdles enterprises face when deploying digital workers at scale: knowing what those agents are actually doing and ensuring they can work securely across corporate systems. The company's Agentforce 3 release introduces a comprehensive "Command Center" that gives executives real-time visibility into AI agent performance, plus native support for emerging interoperability standards that allow agents to connect with hundreds of external business tools without custom coding. The timing reflects surging enterprise demand for AI agents. According to Salesforce data, AI agent usage has jumped 233% in six months, with over 8,000 customers signing up to deploy the technology. Early adopters are seeing measurable returns: Engine reduced customer case handling time by 15%, while 1-800Accountant achieved 70% autonomous resolution of administrative chat requests during peak tax season. "We have hundreds of live implementations, if not thousands, and they're running at scale," said Jayesh Govindarajan, EVP of Salesforce AI, in an exclusive interview with VentureBeat. The company has moved decisively beyond experimental deployments, he noted: "AI agents are no longer experimental. They have really moved deeply into the fabric of the enterprise." "Over the past several months we've listened deeply to our customers and continued our rapid pace of technology innovation," said Adam Evans, executive vice president and general manager of Salesforce AI, in announcing the platform upgrade during a live event Monday. "The result is Agentforce 3, a major leap forward for our platform that brings greater intelligence, higher performance, and more trust and accountability to every Agentforce deployment." How global food giant PepsiCo is leading the enterprise AI agent revolution Among the companies embracing this technology is PepsiCo, which is deploying Agentforce as part of a broader AI-driven transformation of its global operations. In an exclusive interview with VentureBeat, Athina Kanioura, PepsiCo's chief strategy and transformation officer, described the deployment as crucial to the company's evolution in an increasingly complex marketplace. "As a longtime partner of Salesforce, we recognized an opportunity to holistically integrate the way we utilize their platforms across our business -- especially as the customer landscape evolves, trade becomes more complex, and the need to better integrate our data increases," Kanioura told VentureBeat. The food and beverage giant, whose products are consumed over a billion times daily worldwide, sees AI agents as essential for meeting customers "where they are -- and in the ways they want to engage with us," while driving backend efficiency by integrating systems and simplifying processes. PepsiCo's seven-year relationship with Salesforce positioned the company to move quickly on AI agents. "We were excited about how Agentforce could enhance the day-to-day experience for our field sellers - streamlining workflows and surfacing smarter insights in real time," Kanioura explained. The missing piece: Why enterprise AI needs real-time monitoring and control The Command Center represents Salesforce's response to a critical gap in the enterprise AI market. While companies have rushed to deploy AI agents for customer service, sales, and operational tasks, many lack visibility into how those digital workers are performing or impacting business outcomes. Govindarajan described the challenge facing enterprises that have moved beyond pilot programs: "It's one thing to build an AI agent demo, but when you actually build an agentic system and put it in front of your users, there's a different standard." Companies need tools to understand when AI agents are struggling and when to bring humans into the workflow, he explained. "Teams can't see what agents are doing -- or evolve them fast enough," the company acknowledged in its announcement. The new observability platform provides detailed analytics on agent interactions, health monitoring with real-time alerts, and AI-powered recommendations for optimization. The system addresses what Govindarajan calls "day two problems" - the operational challenges that emerge after initial deployment. "You can have multiple agents for multiple personas, and you need to be able to observe how that's actually impacting the task that needs to get done at scale," he said. This includes managing the handoffs between digital agents and human workers when complex decisions or approvals are required. The system captures all agent activity in Salesforce's Data Cloud using the OpenTelemetry standard, enabling integration with existing monitoring tools like Datadog and other enterprise systems. This addresses enterprises' need to incorporate AI agent oversight into their existing operational workflows. Open standards and secure integration: How AI agents connect across enterprise systems Perhaps more significant is Salesforce's embrace of the Model Context Protocol (MCP), an emerging open standard for AI agent interoperability. The platform will include native MCP support, allowing Agentforce agents to connect with any MCP-compliant server without custom development work. "There's generic interoperability, and then there's what we call enterprise-grade interoperability," explained Gary Lerhaupt, VP of product architecture at Salesforce, in an exclusive interview with VentureBeat. "If it's not enterprise grade, it's like sparkling untrusted interop." The key difference, he said, lies in governance and control mechanisms that enterprise customers require. This capability, working alongside an expanded AgentExchange marketplace, gives enterprises access to pre-built integrations with over 30 partners including Amazon Web Services, Box, Google Cloud, IBM, PayPal, and Stripe. Lerhaupt said the company is launching with "north of 20, maybe 25 plus" vetted MCP servers, with partners like PayPal offering invoicing capabilities and Box providing document access through their MCP implementations. "In a world full of AI tools, Agentforce stood out not just for its first-of-a-kind technology but how seamlessly it fit into our technology ecosystem, the way we work and our AI strategy, standards and framework," Kanioura said. Performance boost: Faster AI models and enhanced security for regulated industries Underlying the new features is what Salesforce calls an enhanced "Atlas" architecture designed for enterprise-grade performance and security. The platform now offers 50% lower latency compared to January 2025, response streaming for real-time user experiences, and automatic failover between AI model providers to ensure continuous operation. For regulated industries, Salesforce's approach to hosting AI models directly within its infrastructure addresses critical security concerns. "With Anthropic, the entire stack will be running within Salesforce infrastructure," Govindarajan explained. "The calls are not going out to OpenAI, and traffic will be running within the Salesforce VPC. For regulated industries, that's what we've been working on." Critically for regulated industries, Salesforce now hosts Anthropic's Claude models directly within its infrastructure via Amazon Bedrock, keeping sensitive data within the Salesforce security perimeter. The company plans to add Google's Gemini models later this year, giving enterprises more options for AI model governance. The platform also expands global availability to Canada, the UK, India, Japan, and Brazil, with support for six additional languages including French, German, Spanish, Italian, Japanese, and Portuguese. From zero to AI agent: How pre-built industry actions speed enterprise deployment Recognizing that enterprises need faster returns on AI investments, Salesforce has built over 200 pre-configured industry actions -- with more than 100 added this summer alone. These range from patient scheduling in healthcare to advertising proposal generation in media, designed to help companies deploy functional AI agents quickly rather than building from scratch. The results demonstrate the platform's maturity. Beyond 1-800Accountant's 70% deflection rate during tax season, Govindarajan cited other production deployments: "OpenTable sees 73% of all restaurant web queries handled by agents," and "Grupo Falabella, a Colombian customer service operation using WhatsApp, achieved a 71% reduction in phone call traffic in just three weeks." The company also introduced more flexible pricing, including unlimited usage licenses for employee-facing agents and per-action pricing that scales with actual AI work performed rather than simple conversation volume. The new digital workforce: What enterprise AI adoption means for business operations As enterprises increasingly view AI agents as digital employees rather than simple automation tools, the stakes for getting deployment right have never been higher. Companies that successfully scale AI agents stand to gain significant competitive advantages, while those that struggle with governance and oversight risk operational disruptions. Govindarajan sees fundamental changes in how work gets organized: "New roles are emerging for people who manage a fleet of agents," he said. "A CIO might ask, 'I have seven agents running in my enterprise, what's broadly happening?' But someone running a specific marketing agent has a different lens on the same problem." Looking ahead, Lerhaupt positioned the current moment as transformational: "You had the personal computer, then the Internet, and now it's multi-agent," he said. He described the evolution from single-agent deployments to what he calls "the multi-agent revolution and the ability to plug agents together to do exceedingly complex new types of work." For PepsiCo, the transformation goes beyond efficiency gains. "AI and technology are reshaping enterprise operations in ways that were once unimaginable," Kanioura said. "The work we're doing with Agentforce is one element of PepsiCo's broader transformation as a connected company, paving the way for a more resilient and adaptive future of work." The competitive landscape is intensifying as major technology companies race to establish AI agent platforms. When asked about competition from Microsoft, Google, and Amazon, Govindarajan emphasized Salesforce's integration advantages: "We are able to track the entire cycle of work within the enterprise ecosystem," he said. "We can define flows and interactions in the enterprise, and we've been open and extensible in bringing in your data, your actions, and orchestrating them effectively." The Agentforce 3 platform becomes generally available Monday, with several features including hosted Anthropic models and the full Command Center rolling out through August. But perhaps the most telling sign of the technology's enterprise readiness isn't in the feature list -- it's in the confidence of companies like PepsiCo to bet their digital transformation on AI agents they can finally see, measure, and control.

Salesforce unveils Agentforce 3, its smartest agent platform yet

Also include MCP support and over 100 third-party service connections Salesforce has revealed the next generation of its AI agent platform, promising more visibility and control over your systems going forward. The company says Agentforce 3 marks another big step in the field of digital labor, allowing customers to create and deploy smarter and more powerful agents than ever before. The launch includes a new Command Center tool to give users "complete observability" as well as some major upgrades when it comes to model support, and over 100 new prebuilt industry actions. "With Agentforce, we've unified agents, data, apps, and metadata to create a digital labor platform, helping thousands of companies realize the promise of agentic AI today," said Adam Evans, EVP & GM of Salesforce AI. Salesforce says its new Command Center, part of Agentforce Studio, will offer users much more oversight into the work their AI agents are carrying out on a daily basis. Previously, this had included observing technical issues around the safety and performance of models - however this will now be expanded to cover agent health, performance and outcome optimization. This will allow users to monitor and analyze every interaction carried out by an agent, spotting trends and preventing issues before they happen, and offering natural language support for generating topics, instructions and case studies. It will provide real-time contextual information pertaining to the specific agent's work tasks, and can also offer AI-powered recommendations for tweaks and edits to agents, hopefully making them even more effective. Elsewhere, Agentforce 3 includes built-in support for the Model Context Protocol, providing much greater support for plug-and-play compatibility with a wide range of other agents and services without the need for custom code. Customers will be able to connect to numerous third-party tools and resources, including Amazon Web Services, PayPal, Box, Cisco Systems, Google Cloud, IBM, Notion Labs, Stripe, Teradata and Writer. "Over the past several months, we've listened deeply to our customers and continued our rapid pace of technology innovation," Evans added. "The result is Agentforce 3, a major leap forward for our platform that brings greater intelligence, higher performance, and more trust and accountability to every Agentforce deployment. Agentforce 3 will redefine how humans and AI agents work together -- driving breakthrough levels of productivity, efficiency, and business transformation."

Salesforce launches Agentforce 3 with greater AI agent visibility and connectivity - SiliconANGLE

Salesforce launches Agentforce 3 with greater AI agent visibility and connectivity Salesforce Inc. today announced the launch of Agentforce 3, a major upgrade to its flagship AI product for enterprises with new ways to observe and control artificial intelligence agents on the platform. The Agentforce platform provides companies the ability to build, customize and deploy generative AI agents, which augment the work of employees autonomously. They are goal-oriented pieces of software capable of completing tasks with little or no human supervision. Using the platform, employees across sales, service, marketing and commerce can customize AI "workers" to take action on their behalf using business logic and prebuilt automations. Today, Salesforce announced the launch of a new Command Center that provides complete observability and built-in support for the Model Context Protocol for plug-and-play compatibility with other agents and services. The company also said it is adding over 100 new pre-built industry actions to accelerate the deployment of industry-standard AI agents. As agents perform tasks behind the scenes and collaborate with human workers, there is a growing need for improved visibility. Salesforce offers an observability platform to address technical issues that can affect the safety and performance of models. However, the new Agentforce Command Center goes a step further by unifying agent health, performance, and outcome optimization. Built into Agentforce Studio, which serves as the AI agent customization platform, the command center allows teams to analyze every agent interaction, drill into specific moments and understand trends. It will also display AI-powered recommendations for tagged conversation types to continuously improve Agentforce agents. The command center will act as a single place to understand AI agents changing contextually according to the type of agent that is under display. For example, the metrics and graphs in the dashboard for a product delivery agent will display cancellations, shipments and other metrics. Whereas an online advertising agent will have observability for clickthrough and campaign success metrics. Users will be able to use natural language to generate topics, instructions and test cases right in Studio. Testing Center simulates AI agent behavior at scale with data state injection and AI-driven evaluation, allowing users to stress-test agents before going live. Agentforce 3 now allows AI agents to natively connect to other services, tools and agents by using an open standard called Model Context Protocol, an open standard pioneered by Anthropic PBC. This allows AI agents to connect to services and other AI agents plug-and-play like a "USB-C for AI," without the need for custom code, still governed by existing security policies. Salesforce's integration platform, MuleSoft, can convert any existing application programming interface into an MCP connector, complete with security policies, activity tracing and traffic controls. This will allow teams to orchestrate and govern multi-agent protocols. For example, a user could quickly build an AI agent to use PayPal using natural language instructions to send and receive PayPal invoices using native built-in MCP integration. Using the protocol, no custom code is needed only building the agent in the Agentforce builder interface, testing it and deploying it. Customers will be able to connect to numerous third-party tools and resources including Amazon Web Services Inc., PayPal Inc., Box Inc., Cisco Systems Inc., Google Cloud, IBM Corp., Notion Labs Inc., Stripe Inc., Teradata Corp. and Writer Inc.

Salesforce Launches Agentforce 3 with MCP to Tackle AI Agent Scalability in Enterprises | AIM

"With Agentforce, we've unified agents, data, apps, and metadata to create a digital labour platform." Salesforce has launched Agentforce 3, a significant update to its AI agent platform, aimed at solving key barriers to enterprise-scale adoption: visibility and control. The update introduces the Agentforce Command Centre, a complete observability solution that enables organisations to monitor, measure, and optimise AI agent performance across workflows. "With Agentforce, we've unified agents, data, apps, and metadata to create a digital labour platform, helping thousands of companies realise the promise of agentic AI today," said Adam Evans, EVP and GM of Salesforce AI. "The result is Agentforce 3, a major leap forward for our platform that brings greater intelligence, higher performance, and more trust and accountability to every Agentforce deployment." Agentforce 3 builds on deployments from over 8,000 customers since its October 2024 launch. According to Salesforce, companies like Engine, Grupo Globo, PepsiCo, UChicago Medicine, and 1-800Accountant have already seen measurable outcomes. Grupo Globo reported a 22% increase in subscriber retention, Engine reduced average customer case handle time by 15%, and 1-800Accountant saw autonomous resolution of 70% of administrative chat engagements during the 2025 tax season. Ryan Teeples, CTO at 1-800Accountant, said, "We've established a strong deployment foundation and are focused on launching new agentic experiences and AI automations through Agentforce's newest capabilities. With a high level of observability, we can see what's working, optimise in real time, and scale support with confidence." The new Command Centre provides detailed analytics on latency, escalation frequency, and error rates, while offering real-time alerts and dashboards. It integrates with existing tools like Datadog, Splunk, and Wayfound via OpenTelemetry. It also supports AI-assisted development through Agentforce Studio, allowing teams to simulate agent behaviour at scale and generate topics and test cases using natural language. Agentforce 3 also introduces native support for Model Context Protocol (MCP), enabling seamless interoperability with external systems. MuleSoft connectors allow existing APIs to be converted into MCP-compliant servers, while Heroku offers infrastructure for managing custom MCP servers. This integration supports secure, plug-and-play agent workflows. "Salesforce's open ecosystem approach, especially through its native support for open standards like MCP, will be instrumental in helping us scale our use of AI agents with full confidence," said Mollie Bodensteiner, SVP of operations at Engine. "That level of interoperability has given us the flexibility to accelerate adoption while staying in complete control." The AgentExchange platform has also been expanded to include over 30 partners, including AWS, Google Cloud, Box, Cisco, IBM, PayPal, Stripe, Notion, WRITER, and Teradata. Customers can access a growing catalogue of MCP servers that allow agents to connect with external services. Healthcare institutions are also finding use cases. "AI tools in healthcare must be adaptable to the complex and highly individualised needs of both patients and care teams," Tyler Bauer, SVP for system ambulatory operations at UChicago Medicine, said. "We need to support that goal by automating routine interactions in our patient access centre...which would free up the team's time to focus on sensitive, more involved, or complex needs." According to a forthcoming Slack Workflow Index, AI agent usage has surged 233% over the last six months. Salesforce's updates are positioned as a response to this spike, addressing long-standing gaps in observability, governance, and enterprise readiness. The company stated it will continue rolling out department-specific Command Centre configurations, starting with Service Cloud. These configurations allow supervisors to track agent performance in real time and escalate issues when needed. Agentforce 3 is available now, with expanded global availability and flexible pricing designed to accelerate time-to-value for enterprise customers.

"That moment in AI where people need to be able to try it out and kind of figure it out" - Salesforce AI GM Adam Evans on making things easier as Agentforce 3.0 arrives

A bold statement from Adam Evans, GM of AI at Salesforce, coming roughly nine months after the firm released its Agentforce offering into the wild. As of last (official) count, there are currently some 8,000 deployments of some size or other, with half of those being paying engagements. The other half are those organizations that are still at the 'suck it and see' stage, as Evans notes: We're at this critical inflection point where companies need to really understand what's possible before they can really get into bringing this to scale, which is where we're seeing our customers that started after Dreamforce [when Agentforce was formally announced]. They're moving from pilot to production and scaling. What we want to do is open that up to more and more, whether they're existing customers or new customers. From Dreamforce last year up until today, the aim has been, 'Let's get people up and running with one agent', he adds: Now we're starting to think about, 'How do we get them running with multiple agents working together and orchestrating them together?', and, 'How do we get them to connect to the tools that they already have built?'. We have thousands of customers that have actively deployed agents. They're now on their second, third, fourth, fifth...As they deploy more and more agents, they need to be able to measure their efficacy. They need them to play well with other technologies in their stack. And they need to be able to go even faster as newer and newer use cases come to bear. It has, as noted above, only been around nine months since most users could actually begin to dabble with Agentforce, but the evolutionary phases it's moved through have been rapid. Evans explains: When we launched Agentforce, that was laying the foundation for autonomous agents. And then three months after that, version 2.0 was released, and this is when we brought agents to work more with teams, where people are working in Slack and beyond. Earlier this year, we launched 2DX, and this brought tooling that was necessary for the agent lifecycle, for developers and more to be able to build and test agents. [It was] also bringing agents more embedded into work. Non-conversational kind of background work became possible. We're listening to our customers and we're hearing that they are asking us, 'How do we scale this further?'. So we're asking ourselves the question, 'How do we take agents beyond just a couple departments and bring it across the entire business? What would we need to do this?'. And the answer is that we need to be able to answer the questions, 'What are our agents doing? Are they performing well? Every action, every impact, every workflow, do we have the observability and oversight for these agents to really, truly build out that hybrid workforce?'. All that being so, it must be time for another release. Step forward, Agentforce 3.0, launched yesterday. This comes with a lot of extra bits and pieces, but most notable is the promise of full visibility and observability via a dedicated Command Center within Agentforce Studio. This is pitched as a unified observability layer allowing teams to monitor agent activity, track health metrics, and optimize outcomes in real time. Command Center is an evolution of several things, argues Evans: When we launched Agentforce, what we saw at the beginning was just metrics around uptime, latency, and performance, more of a technical audience for builders. And then in the second phase, we moved into more of following paths and thinking about, 'Is the agent performing the job that I want it to do consistently across time?, and 'How's the experience?'. Now what we're seeing, as we think about the new hybrid workforce [of humans and agents], as agents are moving more into the line of business, as agents start working in a hybrid way with humans, we're thinking about the KPIs that matter more to a line of business. For Sales, maybe it's agents that are single or multiple agents working together to qualify and schedule meetings with sales reps or for Service, being able to do resolution at a higher rate as well. So this is moving from deeper technical to new personas, moving more to the line of business with Command Center. There is also integration with tools like Splunk, Datadog, and OpenTelemetry to ensure compatibility with existing observability stacks. Interoperability is another major development with version 3.0 which comes with built-in support for the Model Context Protocol (MCP) open standard. This enables plug-and-play connections to over 30 partners such as AWS, Google Cloud, Stripe and PayPal. MCP is "all the rage", notes Evans, adding: This is the continued acceleration that we're seeing across the board with AI impacting everything. In this case, it's about how to connect systems and interoperability. I think what's really important is the idea that it isn't now hard to actually create the connectivity. It's how to make sure that you're using it correctly and governing it and being able to put in the instructions and guardrails. It's not just connecting the systems; it's making sure that you're using those connected systems in the way that you want, and you've got the right oversight of how that works. This is basically a commitment to making agents and Agentforce work throughout all systems. It's going to be a more connected environment to get work done. We want to go wall-to-wall to help our customers solve the problems that they need to solve. So you'll see more of that. Whether bringing in products or connecting products, you can expect to see more of that interoperability investment going forward. Going back to those approved adoption numbers, another stat of note is that in the last quarter, some 30% of Agentforce deals reported were expansions. In other words, they came from customers who had dipped their toes in the agentic water and come back for more. That implies a healthy core to build out from, but the big challenge remains how to get the half of users who currently aren't cutting a check. Salesforce has been very careful to set expectations around when Agentforce will make a serious impact on the bottom line, but that clearly has to be the direction of travel. To that end, version 3.0 is pitched both at making it easier to scale initial deployments and for new users to come on board. It comes with more than 100 new pre-built industry actions to map onto various business sectors, from patient scheduling in healthcare to campaign generation in advertising. Salesforce has also introduced more flexible pricing to make it easier to get started, says Evans: When we first came out, Agentforce had a $2 per conversation pricing. Since then, we've added now a usage-based pricing at 10 cents per action. This allows you very flexible pricing that's more geared towards what work the agent's actually doing, the actions and the value that it's creating, so you can deploy agents across all kinds of different work. And of course, when you're looking at productivity, to help your employees, we've introduced an unlimited use license that gives an agent and productivity to your employees at an unlimited capacity at a fixed price. (NB: It's not clear whether this unlimited capacity fixed price has a rate card or is negotiated on a case-by-case basis. I've asked for clarification here.) The overall objective is straightforward, concludes Evans:

Salesforce Agentforce 3 Brings MCP Support, Command Center Partner Opportunities

Solution providers can 'start really thinking about how to build the management capacities of AI agents as we start seeing more of this kind of hybrid workforce,' Salesforce AI Executive Vice President and General Manager Adam Evans said. Salesforce's third version of its Agentforce artificial intelligence agent development platform offers more capabilities that should interest its solution providers, Adam Evans, Salesforce AI executive vice president and general manager, told CRN during a virtual press conference Monday. Native support for Anthropic's Model Context Protocol (MCP), coming in July, and the new Agentforce Command Center, available in August, are important developments for solution providers who work with the San Francisco-based enterprise applications vendor, Evans said. Partners can expect support for Google's Agent2Agent (A2A) Protocol in the future. For now, MCP support is "a commitment to making agents and Agentforce work through all systems," he said. Whether solution providers "are bringing in products or connecting products, you can expect to see more of that interoperability investment going forward today," Evans said. [RELATED: Salesforce Reveals Price Raises, New Payment Packages Amid AI Investments] Salesforce has about 16,000 partners worldwide. Altaf Shaikh, CEO and founder of Framingham, Mass.-based Salesforce solution provider ListEngage, told CRN in an interview that as the vendor's ecosystem keeps going, partners such as himself with 20 years of experience in Salesforce architecture will continue to stand out from the pack. The AI era has enterprises turning to their trusted solution providers to translate the cutting-edge technology to specific use cases and to make sure data is prepared for insights and secured to preserve authorized access, Shaikh said. "AI might disrupt services organizations," he said. "AI might make things easier. We're talking in terms of tying in large enterprises and small enterprises and bringing in all that data within the Salesforce ecosystem. Agentforce lives on top of good data. Good data lives on top of good workflows. Things get messy in an organization." The Agentforce Command Center observability service for optimizing AI agents will help Salesforce solution providers better manage AI agents, set up key performance indicators (KPIs) and present more advisory and consulting work, Evans said. Salesforce solution providers should expect to see more business-function management and reporting tools from the vendor for the AI era. "There's a lot of opportunity for value-add -- specifically for (partners) like SIs -- to come in and start really thinking about how to build the management capacities of AI agents as we start seeing more of this kind of hybrid workforce," he said. Command Center will allow for agent health monitoring, performance measuring and interaction analysis, according to the vendor. The service is built into Agentforce Studio. Users can track agent latency, escalation frequency, error rates, success rates, cost and adoption, plus signals from third-party vendors including Datadog and Cisco Splunk. Command Center can push real-time alerts for unexpected behavior. An Agentforce Studio application is also slated for an August release, according to the vendor. Capabilities in pilot or beta with general availability expected in July include Anthropic Claude models hosted within the Salesforce trust boundary, MuleSoft MCP and A2A support, and Heroku AppLink. As part of Salesforce's partnership with Anthropic, the AI upstart will work with Salesforce to scale Agentforce adoption with Claude for customers in regulated industries. Claude Sonnet model access in Agentforce comes through Amazon Bedrock hosting within the Salesforce trust boundary. Salesforce also plans to allow Google Gemini use in Agentforce "later this year." MuleSoft's new MCP connectors will allow conversion of any application programming interface (API) and integration into an agent-ready asset, according to Salesforce. The asset will have security policies, activity tracing, and traffic controls for orchestrating and governing multi-agent workflows. Heroku Applink promises to simplify deployment, registration, maintenance and connection to custom MCP servers. Already available Agentforce 3 capabilities include 100-plus new pre-built industry actions to hasten time-to-value, managed MCP server hosting through Herok, increased speed and response streaming, web search for Agentforce data libraries, and Federal Risk and Authorization Management Program (FedRAMP) High authorization for Agentforce's public sector customers. Agentforce 3 also adds support for French, Italian, German, Spanish, Japanese and Portuguese. Another 30-plus languages are gaining support "in the coming months," according to Salesforce. Agentforce 3 has also deployed within Canada, the United Kingdom, India, Japan and Brazil.