For two years Agentforce was something an admin decided to switch on. In Winter '27 that decision disappears: if your org has Agentforce access through its licensing, the platform is enabled for you and the toggle in Setup goes away. At the same time the build experience has been replaced — the new Agentforce Builder, powered by Agent Script, turns agent behaviour into something you can version, simulate and review like code. Together these two changes move the question from "should we try Agentforce?" to "who in our org is allowed to build, and under what guardrails?"
Auto-Enablement: What Happens and When
Enablement is not the same as exposure. If Einstein generative AI is on, admins get immediate access to Agentforce Builder — but other users only gain build access when you assign the Manage AI Agents permission, and no agent affects customers until it is activated. Decide those two things deliberately rather than discovering them in an audit.
Agentforce Feature Upgrades in Winter '27
Agentforce auto-enabled in every eligible org
Agentforce is now on by default for orgs that already have access through their SKU, licence or edition — Foundations on Enterprise, Performance, Unlimited and Developer Edition, plus Agentforce 1 Edition. New orgs are enabled at creation; existing orgs were switched on a rolling basis from the first week of September 2026, and the Agentforce toggle on the Agentforce Agents page in Setup is being removed later in Winter '27. Salesforce states there is no additional cost and no billing change.
New Agentforce Builder on Agent Script
Drafting, testing and deployment collapse into one conversational workspace inside the Agentforce Studio app. Multi-view editing (doc-like editor, low-code canvas, script view), one-click simulation with live reasoning traces, and agents that compile to a portable JSON file for versioning and sharing. Agent Script adds code-level predictability — explicit rules, variables and guardrails — on top of natural-language instruction.
Agent Skills, plugins and a skill registry
Over 100 prebuilt skills at launch plus a registry for your own. Plugins bundle skills with MCP connectors, hooks and configuration so the same governed capability can be reused by several agents and across surfaces instead of being rebuilt per use case.
Agentforce Voice and Intelligent Context
Natural, on-brand voice conversations plus Intelligent Context to ground agents in complex unstructured content — contracts, manuals, policies and knowledge — so answers and actions rest on governed enterprise material rather than model recall.
Agentforce Engagement (formerly Agentforce Lead Nurturing)
The lead nurturing agent is renamed and broadened to cover ongoing buyer engagement across the funnel, with the same guardrail and topic model as other Agentforce agents.
Third-party agent orchestration
Native A2A support means Agentforce can orchestrate agents built on AWS, Azure or Google — or run as a subagent where another platform already owns orchestration. Multi-vendor agent estates are now a supported design, not a workaround.
Legacy vs. new agents run side by side
Agents built the new way live in Agentforce Studio; agents created before Winter '27 continue to be managed in Setup. Customisation, deployment and debugging differ between the two, so inventory which of your agents sits where before you plan changes.
Where Agents Land in Sales Cloud
AI-suggested meeting follow-up actions
Einstein Conversation Insights now proposes next steps on the event record in the mobile app — send requested material, update the opportunity, book the follow-up — so post-meeting admin largely disappears.
Post-meeting voice notes
Where a meeting was not transcribed, reps record a voice note in the mobile app; on-device transcription feeds Einstein Conversation Insights and can trigger downstream automation.
Pipeline Forecasting (formerly Forecasting)
Deal risks, methodology scores, contact insights and agent activity are surfaced inside the forecast view, giving sales leaders a health picture rather than a number alone.
Relationship Maps editing
Contact detail such as opportunity role and department can be edited directly from a side panel, keeping buying-group data current without leaving the map.
Call Coaching for video calls
Competencies can be defined and rated against video call records, and with Agentforce for Sales an AI coaching agent can evaluate calls automatically at scale.
Controls You Still Have as an Admin
- Turn off the Einstein setting on the Einstein Setup page to switch off the Agentforce platform entirely.
- Activate or deactivate any individual agent from Agentforce Builder — nothing is customer-facing until you say so.
- Grant the Manage AI Agents permission only to the people you want building agents; admins already have access.
- Scope each agent's data access through its topics, actions and the permissions of its running user.
- Use Agent Script guardrails to constrain what an agent may say, do and commit.
- Review simulation traces before activation, and keep an owner named for every live agent.
KVP's View
Availability is not readiness
An enabled platform tempts teams to build agents on data nobody has cleaned. Duplicate accounts, abandoned flows and undocumented knowledge produce agents that are confidently wrong. Spend the first sprint on data and knowledge, not on prompts.
Publish a build policy before the first agent
Because anyone with Manage AI Agents can now build, you need a short written policy: who may build, which data classes are off limits, what must be reviewed, and who signs off activation. One page is enough — the absence of it is what causes rework.
Treat Agent Script as source code
Portable JSON and script views mean agents belong in version control, with review and a promotion path through sandboxes. Teams that keep agents as untracked config will lose the audit trail they need later.
Inventory legacy agents now
Agents created before Winter '27 stay in Setup with different customisation behaviour. Know which agents sit where before planning migration, or a routine change will behave unexpectedly.
Pick workflows with a measurable clock
The strongest early candidates are workflows with an obvious time or cost metric — case triage, quote assembly, scheduling, post-meeting follow-up. Measure the before-state in the same week you start the build.
A 30-Day Plan for an Auto-Enabled Org
1. Week one — confirm and contain
- Check whether Agentforce is already enabled in production and sandboxes
- Review who currently holds the Manage AI Agents permission
- Confirm no agent is active that you did not approve
2. Weeks two and three — build one agent properly
- Choose a single workflow with a measurable baseline
- Build it in Agentforce Builder using Agent Script guardrails and simulate before activation
- Ground it with Intelligent Context on curated, owned knowledge
3. Week four — govern and decide
- Publish the one-page build and activation policy
- Add agent definitions to version control and your release process
- Report the measured result and pick the next two workflows
Agentforce turned on in your org before you had a plan?
Contact KVP Business Solutions at info@kvpcorp.com for an Agentforce readiness review covering permissions, data grounding, guardrails and your first measurable agent.
Sources: Salesforce Winter '27 Sales release notes (release 264), Salesforce Winter '27 Agentforce release notes, Salesforce Agentforce 360 announcements, and Salesforce Ben on Agentforce auto-enablement in Winter '27.