The Winter ’27 Release went generally available on October 12, and it is the first one where you can read the release notes and feel the ground move under your job description. The wording Salesforce uses is unusually direct: instead of assisting employees one task at a time, agents now run entire workflows — qualifying pipeline, resolving service cases, booking appointments, underwriting risk.

Fifteen production-ready innovations shipped. This is not a preview of a strategy; it is the strategy, deployed. Here is what changed, and what a Salesforce professional should do about it this quarter.

The change in one sentence

For three years, agentic AI has been sold as a better assistant. Winter ’27 sells it as an owner of outcomes. An assistant waits for a request. An owner of outcomes notices the case, plans the resolution, executes the steps, and reports back — on Slack, Teams, voice, and even legacy enterprise systems, grounded in governed CRM data.

Agentforce is now on by default

Something else changed while you were not looking: Salesforce began auto-enabling the Agentforce platform in eligible orgs in early September, and the rollout continues through the Winter ’27 production waves into mid-October. The platform turns itself on. No agent does — every agent stays inactive until someone builds and activates one, no channels get published, and no user permissions or billing change.

But notice what disappeared: “we never turned it on” used to work as an informal governance policy, and that policy no longer exists. The absence of a decision is now a decision, made by Salesforce rather than by you. Three things to do before your org’s upgrade window: find your actual date on the Salesforce Trust maintenance tab, review exactly who holds the Manage AI Agents permission (in many orgs it was handed out by profile years ago and never reviewed), and decide your position in writing — enable-and-govern, or keep the master switch off until a first use case exists.

The six releases worth learning first

Feature What it does Proof point
Adaptive Experiences & Dynamic Plans An ambient agent listens to cases in real time and updates the resolution plan as facts change 4 customers live, including PowerSchool with 550+ users; Service Rep Assistant past 100 customers
Agentic Commerce Search Product discovery and conversion handled by a shopper agent 13% conversion lift, 17% more add-to-cart
Autonomous Scheduling with Voice Voice agent checks availability, picks a technician, completes the booking 15-minute calls compressed to seconds
Agentforce Contact Center CRM, voice, and every channel natively in one place — no third-party contact centre stack Live in 30+ countries
Agent Skills & Plugins Reusable, governed capabilities instead of rebuilding the same logic per agent 100+ prebuilt skills
Slack Code Coding agents work in shared Slack channels with diffs, previews, and human sign-off Available on all Slack plans

Two more items round out the release and matter for anyone in data or integration: Tableau Knowledge ingests structured and unstructured data to build living knowledge graphs that ground agents in governed business context, and Informatica Headless decouples IDMC’s backend from its UI so developers can call governed data management from VS Code, Cursor, Slack, or MCP servers.

Also shipping: agents that coordinate, remember, and improve

Three quieter capabilities may matter more than the headline features: Multi-Agent Orchestration is now generally available — an orchestrator agent hands off to specialist agents declared as connected subagents, each doing what it does best. Third-Party Agent Orchestration coordinates work across Salesforce agents and external agents — including A2A-style interoperability with AWS, Azure, and Google environments. Curated Memory carries relevant context from one interaction to the next, so a customer does not have to repeat themselves and an agent does not restart cold. And the Optimization Agent helps tune agent performance over time, which is the first sign that Salesforce expects agents to need maintenance rather than just deployment.

What this means role by role

Role What to learn first
Administrator Agent Skills and reusable capability design; data spaces and permission boundaries for agents; evaluation cadence
Developer Slack Code workflow and code review discipline; MCP servers; headless patterns instead of screen-driven logic
Consultant Process decomposition — turning a workflow into measurable agent goals and approval gates
Architect Multi-agent orchestration across platforms; where deterministic rules must replace probabilistic reasoning
Data professional Tableau Knowledge graphs, harmonization, and grounding quality — the input every item above depends on

A 30-day plan that actually fits in a calendar

Week 1: Read the Winter ’27 release notes for your own cloud only. Write down the three features that touch something you already own.

Week 2: Pick one existing automation in your org that a human must currently babysit. Sketch what it would look like as an agent-owned workflow, including the approval gate.

Week 3: Build it in a sandbox using Agent Skills rather than from scratch, and run the Agentforce Testing Center against it.

Week 4: Present the results to your team — including what broke and where the agent needed a human. The people who can narrate a failure honestly are the ones who get trusted with the next workflow.

The feature list is long, and most of it will not touch your org this quarter. What will touch your org is the expectation, set by this release, that workflows run themselves. Being the person who knows where the guardrails belong is the most durable position in that shift. Our Advanced Administrator practice tests cover the automation, sharing, and data-management domains that every one of these workflow ownership stories runs on.