The short version
- AIforce, announced at Dreamforce on September 15, 2026, is a live interface layer that lets people and AI agents use Salesforce data, workflows, and permissions from outside Salesforce, inside Claude, Slack, or the Lightning search bar.
- It launches with three surfaces: [Claudeforce](/blog/claudeforce) (Salesforce in Claude), [Slackforce](/blog/slackforce) (Salesforce in Slack), and [Agentforce Coworker](/blog/agentforce-coworker) (an AI teammate in Lightning).
- Every request runs on your existing permissions, and Salesforce promises Zero Data Retention by the model provider. There is no migration and no new permission model to build.
- The thesis, in Patrick Stokes’ words, is that the value of Salesforce "has never been in the UI." That is directionally right, but "AI replaces the UI" is a keynote line, not a 2026 reality.
- Because AIforce inherits your permissions rather than fixing them, the single highest-value move before you touch it is a permissions and data-model audit. Loose access becomes AI-speed loose access.
Salesforce Spent 26 Years Teaching You to Log In. Now It Says Stop.
On September 15, 2026, at Dreamforce, Salesforce stood on stage and argued that people should stop logging into Salesforce. The announcement is called AIforce, and Salesforce describes it as "a live interface layer that brings the full power of Salesforce to wherever people and agents work." It was the headline of a packed three-day keynote, which we cover end to end in our Dreamforce 2026 recap.
In plain terms: the data, workflows, business logic, permissions, and governance that live inside your org get pushed out to the tools your team already has open. A worker who has never touched a Salesforce dashboard can ask a question, update a record, or trigger a workflow from inside Claude, Slack, or a search bar. As CEO Marc Benioff put it, "AI is creating an interface revolution."
Patrick Stokes, Salesforce’s President of Applications and Marketing, was blunter: 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 and CTO Parker Harris, who built the Lightning interface, went furthest: "Why should you ever log into Salesforce again? Maybe you never will."
The Platform Without the Front Door
Under the keynote language, the mechanism is specific. AIforce runs on the Headless Toolkit, an architecture that exposes the platform through the Model Context Protocol (MCP), APIs, plug-ins, and skills, so an external AI can attach to Salesforce data and act on it. Two things sit behind it: Data 360 harmonizes your data, metadata, and memory so an agent understands the business, and Customer 360 supplies the logic, processes, permissions, and actions to get work done.
The part that matters most for anyone responsible for security: AIforce says it inherits your controls rather than rebuilding them. "Every agent sees only what the person asking can see, and every action routes back through Salesforce." Business data answers the request and, per Salesforce, "is not retained by the model provider." There is no new permissions model and no migration required to switch it on.
The Three Doors
AIforce launches with three surfaces, at different stages of readiness.
[Claudeforce](/blog/claudeforce) is Salesforce inside Anthropic’s Claude, delivered as a prebuilt Salesforce MCP server with 37 ready-to-use sales skills, plus a development plug-in for Claude Code with more than 40 skills. It is in beta. If you are weighing it against your existing agents, we compare the two in Claudeforce vs Agentforce.
[Slackforce](/blog/slackforce) is Salesforce inside Slack. Its headline feature, Slackforce Surfaces, turns a plain-language request into a live, interactive dashboard or report built from your Salesforce and Slack context, one that stays connected and refreshes on its own.
[Agentforce Coworker](/blog/agentforce-coworker) is an AI teammate in the Lightning search bar that reasons across accounts, activity, and history and takes action. It is the one you can use today, and its proof point is real: Salesforce says 100,000 users activated it within its first 35 days.
Koa, Quietly, Underneath
Alongside AIforce, Salesforce announced [Koa](/blog/salesforce-koa), its first CRM-specific reasoning model, built with NVIDIA. Koa is not a surface you interact with. It sits inside Agentforce and gets selected automatically to handle high-volume, well-scoped CRM actions at lower cost, with frontier models like Claude reserved for harder judgment. The economics of Koa are arguably the more important story of the week, and worth understanding on their own.
The Proof, and the Asterisks
The numbers Salesforce put forward are, for a launch, unusually specific. Fulton Bank went "from zero to more than 20 use cases live in production, supporting approximately 3,000 users" in weeks with Agentforce Coworker. Coworker hit 100,000 activations in 35 days. Engine, which handles over 800,000 customer inquiries a year, used Slackforce so anyone could build a working interface from existing data.
They are genuine, and worth reading carefully. Fulton Bank is a bank, not a mid-market business, and the flagship references are large enterprises. Two of the three surfaces are in beta. "Instantly available, no migration" is Salesforce’s framing, and a launch-week activation count measures curiosity as much as value.
The Honest Read
The thesis is directionally right. The durable value of a Salesforce org was never the screens, it was the encoded business logic, the data model, and the permissions. Letting AI front-end that from anywhere is a real shift, not a rebrand.
But "AI replaces the UI" is a keynote line, not a 2026 reality. More importantly, AIforce inherits your permissions rather than fixing them. The open question, as one trade outlet put it, is whether those controls hold up "without quietly weakening the controls that made CRM trustworthy in the first place." If your sharing model is loose today, AIforce turns loose permissions into loose, conversational, AI-speed access to the same data.
And value is proportional to hygiene. AIforce surfaces "the way our customers encode their business." An org with a clean data model, tight permissions, and well-designed agents gets close to the demo. An org with duplicate fields and stale automations gets a confident interface on top of a mess.
What a Mid-Market Team Should Do Now
Do not rip anything out. The UI is not going away, and nothing here requires a migration. Treat AIforce as an addition.
Audit your permissions and sharing model first. This is the single highest-leverage move, because AIforce inherits exactly what you have. Over-permissioning that was survivable when access meant clicking through screens is a real exposure when an agent can read across hundreds of records on request.
Pilot one surface, narrowly. Agentforce Coworker in Lightning is the lowest-friction start; Salesforce-in-Claude suits a sales team already living in an AI tool. Pick one, scope it, measure it.
The companies that win with AIforce will be the ones that did the boring work first: clean data, tight permissions, well-designed agents. That is the senior implementation work our Agentforce practice and Claudeforce practice exist to do, and a free scoping call is the fastest way to find out where you stand.