What's Next in Salesforce AI: Key Innovations from Dreamforce 2026

Enterprise AI is changing roles, from "an assistant that answers questions" to "a team that gets work done."

AI is no longer limited to finding information, summarizing documents or answering questions. It is starting to work alongside people across the whole process, from taking on a task and planning it to executing, tracking results and closing it out, whether that means nurturing prospects or handling service cases.

Let's look at what Salesforce, the world's #1 CRM, unveiled at Dreamforce 2026 (September 15–17, San Francisco), revealing how AI and CRM are set to work together in new ways.

Marc Benioff, Chair and CEO of Salesforce, summed up the idea on stage in one line: "The AI you want. The trust you need." In other words, organizations can choose the AI that suits them without giving up control over their data.

As a Salesforce Partner, BE8 has rounded up three key innovations from the event:

• AIforce: brings Salesforce into the tools you already work in, including Claude, Slack and Microsoft Teams
• Koa: Salesforce's AI model, built specifically to understand and reason through CRM work
• Prebuilt agents for Agentforce: AI agents ready to work from day one, from finding customers to supporting and resolving service issues

Here's what each one does, and what it can mean for your organization.

1. AIforce: the live interface for Salesforce, anywhere

Work is moving into AI-first, conversational experiences. Instead of logging in and clicking through screens, people now ask an AI for what they need. Salesforce calls AIforce "the live interface for Salesforce": it brings the workflows, data and business context that make Salesforce useful into the places where work now happens.

In the AIforce Keynote, Salesforce described it as bringing "all of Salesforce's capabilities to any AI, any interface, any agent, wherever work happens." That means your Data 360 foundation, your Customer 360 business context and your Agentforce digital workforce, all grounded in Salesforce's trust and governance.

AIforce shows up across the surfaces teams already use:

• Claudeforce (Salesforce in Claude): co-developed with Anthropic, set up in a few clicks, and in beta for all customers from Dreamforce week. It comes with MCP and 37 built-in skills, so sales and business operations users can get started without a technical setup.
• Slack and Microsoft Teams: teams work with Salesforce data through Slackbot and in Teams, without switching tools.
• Agentforce Coworker: a conversational interface for all of Salesforce inside Lightning.

In the keynote demo, an events company ran an entire festival day across Claude, Slack and Agentforce Coworker, fixing thousands of at-risk VIP records, all from one source of truth in Salesforce.

For builders, a headless toolkit sits underneath AIforce: APIs and MCP servers, skills and plugins, UI widgets, plus identity and observability. Everything a company has built in Salesforce over the years, from custom objects to permissions and business rules, becomes available to agents. A new agentic identity gives every agent its own identity, so its actions can be governed, tracked and audited.

As Patrick Stokes, Salesforce's President of Applications and Marketing, said in the Main Keynote: "Your data is your data. You're not training any other models when you're using our products."

2. Koa: Salesforce’s First CRM Reasoning Model for Agentforce

One message Salesforce heard from customers everywhere: they want a choice of AI, picking the model with the best price-performance for the outcome they need. In response, Rohan Kumar, Salesforce's President of Platform and Engineering, introduced Koa, "Salesforce's first CRM reasoning model for Customer 360."

• Built on CRM know-how: Koa brings together 27 years of CRM product innovation and more than a decade of AI and data research.
• Made for long-running agents: it supports agents that carry out many tasks to reach a complex outcome, such as a sales agent finding the next best step on a valuable deal, or a service agent working through a complicated support case.
• Trusted by design: Koa was trained only on synthetic data, building on NVIDIA Nemotron. In Kumar's words: "Not a single byte of customer data was used."

In the Data 360 Keynote, Salesforce placed Koa inside its Enterprise AI Harness under "trusted models": Koa sits alongside frontier models such as Claude, so companies can choose the right model for each job and keep costs under control.

Koa's synthetic training data reflects real business processes across 14+ industries. On Salesforce's own CRM Bench testing, Koa compared with general-purpose models is:

• 11% more precise at selecting the right action
• 2.1x more reliable at recalling customer context
• 15% better at keeping conversation context
• 3x fewer errors on CRM tasks

Koa runs entirely within Salesforce's trust boundary and is hosted at "temperature 0" for consistent results. Admins can select it as a managed model for the whole org in Agentforce Setup, or set it for individual agents. Pilot customers include Formula 1, Xero, UChicago Medicine and 1-800Accountant. Koa is in pilot now, with general availability expected in U.S. regions in Winter 2026.

Learn more about Koa

3. Prebuilt agents for Agentforce: job-ready from day one

If AIforce is the new interface for work, Agentforce is the agent platform behind it. At Dreamforce, Salesforce showcased out-of-the-box agents that sit on top of Agentforce, which business users, not just IT, can implement and deploy.

The common thread: each agent is opinionated and ready to work out of the box, then adapts to your data, metrics and processes.

From the Dreamforce stage to your business

Each of the three innovations helps organizations in a different way:

• Faster work, in the tools your teams already use: AIforce lets people use Salesforce data and workflows from Claude, Slack or Teams, with no screen switching or new system to learn.
• More accurate AI decisions: Koa is designed to understand how deals, customers and cases are structured, to help reduce mistakes when AI updates records or chooses the next step.
• Faster time to value: prebuilt agents start with clear jobs, such as finding leads, answering customers or handling IT and HR requests, so you don't build from zero.

Whether it's AIforce, Koa or prebuilt agents, they all rest on the same principle: AI is only as accurate as the data and business rules behind it. Organizations that get this foundation ready will be the first to turn innovations from the Dreamforce stage into real business results.

At BE8, we believe putting AI to work starts with finding a use case that fits the business and can be measured. From there, we match the right Salesforce capabilities to that need, building on the data and systems the organization already has, so AI delivers results you can see.

Before you start: 3 questions to ask

1. Is your data ready? How complete and accurate is your customer data, and how well is it connected across systems?
2. Where should you start? Pick repetitive work with clear metrics, such as self-service or lead follow-up.
3. Is your governance ready? Do your permissions and governance cover agents: who approves what, and how can actions be audited?

Start with BE8: AI you can measure

BE8 combines a deep understanding of Thai business with expertise in Salesforce technology, and has experience delivering Salesforce projects for organizations across many industries.

Talk to BE8's experts to find the first use case that fits your business and can show results quickly. Click

Source: Salesforce Dreamforce 2026 Media Resources, Salesforce Koa Announcement