The short version
- Dreamforce 2026 ran September 15 to 17 in San Francisco, and the whole event pushed one idea: AI replaces the UI. The marquee launch, [AIforce](/blog/aiforce), puts your Salesforce data, logic, and permissions inside Claude, Slack, and the Lightning search bar.
- Salesforce shipped its first own model, [Koa](/blog/salesforce-koa), a CRM reasoning model built with NVIDIA, and made cost, routing cheap Koa versus frontier models, the real story.
- Agentforce got seven named "job-ready" agents (Casey, Paige, Carter, Marshall, Piper, Fin, and Hunter) and a long-horizon runtime that lets an agent pursue multi-week objectives, not just single tasks.
- Big partnerships landed: Google Cloud (Hyperforce and Gemini Enterprise), AWS (Amazon Quick and Bedrock), NVIDIA (Koa), and Siemens. Keynote guests included Anthropic’s Dario Amodei and NVIDIA’s Jensen Huang.
- Real proof and real caveats: Salesforce cited 7 billion Agentic Work Units and marquee customers like Fulton Bank, Adecco, and Air India, but most of AIforce is still beta and the Koa accuracy claim rests on Salesforce’s own benchmark.
One Idea Ran Through the Whole Event: AI Replaces the UI
Dreamforce 2026 ran from September 15 to 17 at the Moscone Center in San Francisco, and for all the products on stage, it made a single argument: you should stop logging into Salesforce. Marc Benioff called it "an interface revolution." Patrick Stokes, President of Applications and Marketing, put the thesis plainly: the value "has never been in the UI. It’s been in the platform that stores the way our customers encode their business." Co-founder Parker Harris, who built Lightning, went furthest: "why should you ever log into Salesforce again?"
The keynote was unusually crowded with partners. Anthropic CEO Dario Amodei, NVIDIA CEO Jensen Huang, Siemens CEO Roland Busch, and Adecco Group CEO Denis Machuel all appeared. That guest list is the strategy in miniature: Salesforce is positioning itself as the trusted layer of context and governance underneath everyone else’s AI, rather than a destination you visit. Here is everything that mattered, and an honest read on what actually changed.
The Marquee Launch: AIforce and Its Three Surfaces
[AIforce](/blog/aiforce) is the product expression of "AI replaces the UI." It is a live interface layer that takes the data, workflows, business logic, permissions, and governance inside your org and pushes them out to whatever tool your team already uses. Every request runs on your existing permissions, and Salesforce promises Zero Data Retention by the model provider, with no migration required.
It launched with three surfaces, and they are worth understanding individually:
- [Claudeforce](/blog/claudeforce): Salesforce inside Anthropic’s Claude, with 37 prebuilt sales skills, delivered through a Salesforce MCP server. In beta.
- [Slackforce](/blog/slackforce): Salesforce inside Slack, including Slackforce Surfaces, which builds a live, connected dashboard or report from a plain-language request. In beta.
- [Agentforce Coworker](/blog/agentforce-coworker): an AI teammate in the Lightning search bar that reasons across your accounts and takes action. Available now, with 100,000 users activated in its first 35 days.
Each of those has its own deep-dive above. If you read one, read the AIforce explainer first, because it frames the other three.
Salesforce Built Its Own Model: Koa
The most strategically interesting announcement was [Koa](/blog/salesforce-koa), Salesforce’s first CRM-specific reasoning model, built with NVIDIA by post-training NVIDIA’s Nemotron on synthetic data modeled on nearly three decades of CRM work. No customer data was used to train it. Salesforce says Koa "matches or exceeds leading model performance on CRM actions with three times fewer errors" on its own CRM Benchmark.
The real story is cost. Inside Agentforce, an AI gateway routes routine, well-scoped work to the cheap, CRM-tuned Koa and reserves frontier models like Claude for genuine judgment. At Salesforce’s scale, that routing changes the unit economics of running agents. Koa is in pilot now, with general availability targeted for winter 2026 in US regions.
Seven Job-Ready Agents and a Long-Horizon Runtime
Beyond the interface story, Agentforce itself grew up. Salesforce introduced seven named "job-ready" agents, pre-built for specific functions so a team does not start from a blank canvas:
- Casey, the help agent, resolves service issues across voice, SMS, WhatsApp, and web chat.
- Paige, the employee agent, handles IT and HR requests in Slack and portals.
- Carter, the shopper agent, assists product discovery and in-chat checkout.
- Marshall, the supply-chain agent, orchestrates back-office work with audit records.
- Piper, the pipeline-generation agent, engages and qualifies inbound leads.
- Fin, the customer agent, resolves complex multi-channel workflows.
- Hunter, the outbound-sales agent, works pipelines from research to outreach. In pilot, with general availability planned for November 2026.
Most are generally available now. The bigger shift underneath them is the long-horizon runtime: agents can now pursue multi-week objectives, not just single tasks. Ask Hunter to rescue an at-risk deal and it "builds a plan, and determines the tasks, tools, and context required" within guardrails. Salesforce also shipped Multi-Agent Orchestration (generally available) for routing work across agent teams, and an Agent Optimizer for building and testing agents, slated for October 2026. If you are new to the platform, our getting started with Agentforce guide is the on-ramp, and Claudeforce vs Agentforce clears up how the pieces relate.
The Plumbing: Data 360, Customer 360, and Headless 360
None of the flashy surfaces work without the foundation, and Salesforce spent real keynote time on it. Data 360 harmonizes your data, metadata, and memory so an agent understands the business. Customer 360 supplies the logic, processes, and permissions. And [Headless 360](/blog/what-is-headless-360-complete-guide), the MCP-based architecture introduced earlier in 2026, is what actually lets agents in Claude, ChatGPT, and Cursor invoke Salesforce capabilities directly. AIforce is the polished, packaged version of Headless 360, which is a larger and more technical surface underneath it.
The Partnerships: Google, AWS, NVIDIA, Siemens
Dreamforce 2026 was as much about alliances as products:
- Google Cloud: Hyperforce, Salesforce’s infrastructure platform, is coming to Google Cloud, with customers moving in Q4 2026 and North American general availability in November 2026. Gemini Enterprise can connect to Salesforce capabilities without custom integration.
- AWS: Salesforce business context comes to Amazon Quick, AWS agents arrive in Slack, and Bedrock provides model access, including the containerized path that keeps Claude inside the Salesforce trust boundary for regulated firms.
- NVIDIA: co-built the Koa model on Nemotron, and powers the new Missionforce work for the public sector.
- Siemens: connected Agentforce to Teamcenter service lifecycle management, and deployed an inbound-sales agent across 132 countries, turning roughly 2,500 unqualified leads a month into 100% engaged.
The Numbers, and the Proof
Salesforce backed the story with scale figures: 7 billion Agentic Work Units delivered across Agentforce and Slack (3.2 billion in a single quarter), a Trailblazer community running 12 million applications and 9.6 billion API calls a day, 156 million lines of AI-generated code, and 151 million Trailhead badges.
The customer proof was specific. Fulton Bank went from zero to more than 20 use cases and roughly 3,000 users in weeks with Agentforce Coworker. Adecco is rolling Coworker to 27,000 employees across 40-plus countries. Engine resolves half its chat inquiries fully with AI, and Salesforce named a major new AI agreement with Air India. Koa’s named pilots include 1-800Accountant, Baxter Credit Union, Engine, Formula 1, UChicago Medicine, and Xero. Impressive, and worth reading with the asterisk that the flagship references are large enterprises, not mid-market proof yet.
Beyond Product: A $27 Million Education Commitment
Salesforce paired the product news with philanthropy: a $27 million education commitment, $18 million to public school districts in San Francisco, Oakland, New York City, Chicago, and Indianapolis, and $9 million to educator and career organizations. It sits inside a broader roughly $970 million in grants the company has committed over its lifetime.
The Honest Read: What Actually Changed
Strip away the keynote language and one real shift remains: Salesforce has decided its future is as the governed layer of context underneath other people’s AI, not as an application you open. That is a coherent, defensible bet, and the interface layer is a genuine architectural move, not a rebrand.
But "AI replaces the UI" is a thesis, not a 2026 reality. Two of the three AIforce surfaces are in beta, Koa is in pilot, and several Agentforce features are October or November roadmap. The accuracy claims for Koa rest on a benchmark Salesforce built. And the whole model inherits your permissions rather than fixing them, so the open question is whether governance holds up across external surfaces without quietly weakening the controls that made CRM trustworthy. The value of all of it is capped by the cleanliness of your data model and the quality of the agents you have built.
What a Mid-Market Team Should Do
You do not need to react to a keynote. You need to do the unglamorous work that makes any of this pay off. Three moves:
Audit your permissions and data model first. AIforce and Coworker inherit exactly what you have, so an access and data-hygiene pass is the highest-leverage thing you can do before turning anything on.
Pilot one surface, narrowly. Agentforce Coworker in Lightning is the lowest-friction start, and it is available now. Scope one use case, ship it with a human gate, and measure it.
Build a couple of good agents. The interface layer is only as useful as the agents behind it. That, plus the permissions and data work, is exactly what our Agentforce and Claudeforce practices are built for, and a free scoping call is the fastest way to turn a Dreamforce headline into a plan.