Dreamforce 2026: Salesforce Takes Trapped Enterprise Value to Every Interface with AIforce

Everyone wants to know who will win AI — which model, which company, which one is smartest. Salesforce opened Dreamforce 2026 by saying that is the wrong question. The future is not about picking a single intelligence, but about tapping the finest intelligence available globally and selecting the most fitting one for the task ahead. That shifts focus from model rivalry to context-aware orchestration. That framing set the logic for every announcement that followed: not a model race, but an interface race grounded in business context.

The Trapped-Value Problem: Models Know the World, Not Your World

Frontier models may know almost everything about the world, but they do not know your world — your customers, business, pipeline, inventory, service history, contacts, permissions, processes, conversations, transactions, workflows and business rules. Salesforce has spent 27 years accumulating that knowledge and hammered the point on stage: before you can act, you have to truly understand context. When data is connected you get a foundation and a 360-degree view of the customer; that is where decisions become action across sales, service, marketing and commerce, with precision inside applications you already trust.

On that foundation the company puts its system of agency: Agentforce, where intelligence is multiplied. Agents operate with the entire business context behind them — constructed under governance, rolled out with controls, monitored, refined and trusted. The roll call showed scale: Piper for qualification, Hunter generating $500 million in pipeline in a single quarter, Casey answering 5 million service cases over 18 months, Page for ITSM and HR, Marshall for supply chain, and Fin — the number one service agent — live on help.salesforce.com after a 12-day deployment. Different faces of the same architecture.

Four Layers and a Zero Retention Promise

The keynote framed the architecture in four layers. One, Data 360: the data layer that integrates, federates and harmonizes everything into a single source of truth. Two, the application and semantic intelligence of Customer 360: built metadata-first for more than two decades, so permissions, business logic and processes are ready for agents from day one. Three, the agentic layer matured over three years. Four, the new interface layer: AIforce. With headless apps, the full enterprise knowledge recomposes in that layer.

The interface-revolution thesis was placed in history: DOS to GUI to web to mobile and now to AI interfaces. The difference: these interfaces are dynamic, intelligent, composable and alive. They do not just display information; they help you administer, build and operate applications under governance. The trust base for that is Zero Data Retention, first announced three years ago and stressed as audited, tested and repeatedly probed by security teams: your data is your data, not used to train any other model, not retained by the model provider.

AIforce is the payoff: a live interface layer that brings the full power of Salesforce to wherever people and agents work. Until now knowledge was locked behind opening the app and navigating menus, limiting who could use it and how much insight could be seen at once. With AIforce, Salesforce comes to you — in Claude, Slack, Lightning or wherever you choose to work. Agents can read across hundreds of records, pull in connected systems and reason and act with a complete view at a speed and scale no human can match.

Five Properties: Intelligent, Secure, Composable, Zero Retention, Works With What You Have

Salesforce defined AIforce with five properties. One, intelligent and dynamic: agents reason across all data, logic and conversations at once and surface insights instead of lists to sort. Two, secure and governed: every request runs through existing permissions and business rules, so an agent sees only what the asker can see and every action routes through Salesforce. Three, composable: anyone can build their own interface, agent or view simply by describing it, with MCPs, APIs, plug-ins and skills for teams that want more. Four, Zero Data Retention: business data answers the question at hand and is not kept by the model provider. Five, works with what you have: no new permission model, migration or custom integration — admins connect once and teams have access on day one.

The meaning for Trailblazers is outsized. Celebrating its 20th anniversary at Dreamforce, the community now spans 23 million members across 90 countries and 700 groups, with 12 million apps built, 9.6 billion API calls per day, 156 million lines of AI-generated code and more than 151 million Trailhead badges earned. AIforce gives every Trailblazer the entire architecture at hand to turn ideas into reality at unprecedented scale and speed — admins, developers, architects and builders alike.

Three Flavors: Claudeforce, Slackforce and Coworker

AIforce launches in three flavors — Claudeforce, Slackforce and Agentforce Coworker — with more to come and an SDK to build your own AIforce apps. Claudeforce deepens the Salesforce-Anthropic partnership as the number one AI meets the number one CRM. As a prebuilt MCP server inside Claude, Salesforce in Claude packs 37 prebuilt sales skills covering prospecting to pipeline hygiene for instant value; analytics with Tableau plus service, marketing, commerce and industry skills will follow. For builders, the Claude Code plugin brings more than 40 skills, the broader skills library on GitHub and specialized sub-plugins that load dynamically to take on development tasks. Piloted by Deloitte, GitLab and Legora, it is now in beta for all customers.

Slackforce builds on the idea that Slack is where AI works. Acquired six years ago and now the fastest-growing product, Slack now runs the same AIforce interface natively. Slackforce Surfaces lets users pull live context from Salesforce, Slack and other tools into an interactive, live interface the whole team can filter, explore, comment on and act on together in real time. Slackbot is the out-of-the-box personal assistant that reasons across conversational context in Slack plus the semantic intelligence and governed actions in Salesforce — surfacing quiet accounts, diagnosing why from support cases and threads, reassigning owners, creating follow-up tasks and drafting win-back emails right from Slack. Slack CRM connects every conversation, user and update to Salesforce: spin up an account, log call notes, update records — all by prompt from Slack without opening a separate CRM tool.

Agentforce Coworker is the AI teammate that lives inside Salesforce on every device. It works within business rules, permissions, security and governance, calling specialized Agentforce agents already deployed. Together the three show the same live interface recomposed across surfaces: what was recomposed into browsers and iOS now recomposes into AI interfaces like Cowork and Slack and Lightning.

Live on Stage: How Piper and Marshall Worked for Siemens

Siemens made the theory tangible. In act one, Piper stepped in. When a visitor asked about the commercial model for a distributor of our size, Piper recognized buyer intent, combined it with account context and decided it was time to bring in a seller, locating aligned account executive Ryan Smith inside Salesforce and creating a pipeline booking in a few clicks. No forms, no waiting — quality pipeline generated from a conversation. That qualification flow is already live on informatica.com and salesforce.com; Hunter and Casey's numbers are part of the same scale story.

Act two moved to a completely different function: supplier onboarding, where for Siemens a single supplier can take days and a single disruption can cascade into production delays costing millions. Marshall is designed to shrink that from days to hours. You start in Slack but the agent will find you in email or Microsoft Teams. In the background Marshall collects information, reviews documents and pulls in a human when review is needed. Watching it, most steps fly by; the sticking point is creating the supplier inside SAP — exactly where, the stage stressed, a majority of AI projects hit the wall in back-end systems.

Marshall must first learn. The team gives it an SAP sandbox — a classroom, a safe learning environment without touching production. It learns where to click, which fields are required and, crucially, reasons with LLM intelligence about the business rules behind them, for example that a US supplier needs an additional form. The line on stage was sharp: what takes a new employee weeks to learn, Marshall learned in 90 minutes. It then packages those learnings into a library of trusted actions where reasoning becomes deterministic execution — not improvisation, but the same reliable steps every time.

The test put Marshall to work with demo driver Gus completely hands-free, sipping coffee while Marshall autonomously updated every field inside SAP. Repeatable work that must be done right every time is where agents shine. Back in Slack employees simply see the supplier created, onboarded and activated. If something inside SAP changes, Marshall returns to the classroom to relearn and adapt — trusted process automation. The Siemens takeaway was explicit: Piper for pipeline, Marshall for supply chain, mesma Agentforce digital workforce scaling across the business.

Koa: Salesforce's First Reasoning Model

The technical headline is Koa, Salesforce's first CRM reasoning model built on NVIDIA Nemotron 3 Super and post-trained on a proprietary synthetic dataset modeled on nearly three decades of CRM deployments. No customer data was used; scenarios simulated real enterprise workflows across 14+ industries — manufacturing, financial services, healthcare, travel — pairing personas with tasks and mapping the tool-call sequences needed to complete them. Training applied Supervised Fine-Tuning and reinforcement learning with Group Relative Policy Optimization using NVIDIA NeMo RL, NeMo Gym and NeMo AutoModel. On Salesforce's CRM Benchmark — updating an opportunity, routing a case, scheduling a follow-up — Koa already matches or exceeds leading models with 3x fewer errors, with Salesforce controlling the weights and running inference inside its own trust boundary.

Koa is already in use inside Salesforce as a Slack agent that finds information and completes everyday tasks, and is moving into customer pilots with 1-800Accountant, Baxter Credit Union, Engine, Formula 1, UChicago Medicine and Xero. The CEO quotes converge: accounting needs step-by-step reasoning through tax rules and customer specifics, financial services need to reason across goals and policies, business travel needs precise multi-step problem solving, healthcare needs coordination across longer workflows. Koa now powers those Agentforce use cases in pilots, generally available to select pilot customers with GA expected Winter 2026 in US regions.

Alongside it comes the Missionforce expansion. Bringing NVIDIA open models and accelerated computing into Missionforce, Salesforce gives government and regulated organizations control over model, data and deployment — private clouds, classified networks and fully air-gapped systems. Post-trained NVIDIA models will power Missionforce Operations, digitizing procurement, supplier management and logistics, trained on an organization's operational data and terminology to reason through back-office processes inside the customer's environment. Missionforce Operations is generally available now in US regions; post-trained NVIDIA models arrive for select customers in October 2026. Coverage from SiliconANGLE and Pulse2 adds Google Cloud, AWS and Siemens partnership expansions and a $27 million commitment to education at Dreamforce 2026.

Beta, Pricing and a Human-Centric Close

Pricing and packaging was handled the Dreamforce way: everything is in open beta for now. Find the plugin on the AppExchange, submit the form and org ID, and Salesforce will enable you that day. The stage demo was built in six to eight minutes to a working interface, then iterated over days adding deliberately silly features like sad trombones and a 3D Parker in a lightning suit — skeleton in minutes, personalization over days.

The keynote also closed a narrative that has swirled for months: the SaaSpocalypse. Benioff called it crazy nonsense and reframed it as not the end of software, but the end of software that makes humans do all the work. Numbers were laid out for confidence: $46.4 billion in revenue guidance in the second quarter, another $1 billion in free cash flow, 14% CRPO growth, the lowest attrition and longest contracts ever. The live proof point was operating leaders such as new COO Miguel Milano running the business on Claudeforce and Slack in real time, presented as live usage rather than a pilot.

The close returned to a human frame: business reinvention, measuring outcomes, bringing people along. If AI is to succeed, people cannot be an afterthought, and creating human-centric AI is a shared responsibility. A brief on-stage conversation on behalf of Odeco and a thank-you for leadership led to the call to action: start your agentic enterprise journey. Dreamforce, the message went, is where everybody transforms — and the shift from a few pilots to everyone seeing their systems change through a live interface is meant to make that transformation visible at last.