What changed with Winter '27
Starting with the Winter '27 release, Salesforce turns Agentforce on by default in every org that has access to it through its licenses and edition. New orgs get it on day one. For existing orgs, the rollout started arriving in waves in the first week of September 2026 (Salesforce Ben). Salesforce's official Winter '27 developer guide confirms it: the Agentforce platform is now turned on by default (Salesforce Developers).
It covers orgs running Lightning Experience on Foundations in the Enterprise, Performance, Unlimited and Developer editions, plus orgs with Agentforce 1 Edition (Salesforce Ben).
The part people are debating isn't the switch itself. It's that Salesforce is removing the Setup option that let admins turn Agentforce on or off. From then on, the question is no longer whether it's on, but who uses it and for what.
In one line
Agentforce is already available in your org, even if nobody asked for it. Being on doesn’t cost anything, but it puts the tools within reach of anyone who has the permissions.
On is not the same as running
According to Salesforce, the default enablement carries no additional cost and no billing changes, and the goal is to let teams build, test and deploy agents "with one less step" (Salesforce Ben). Three clarifications:
- No agent starts serving customers on its own. Each agent is activated or deactivated individually in Agentforce Builder.
- Consumption starts when agents work. The actions of an active agent are paid for with credits. We explain the model in Agentforce pricing.
- Admins get in right away. If Einstein generative AI is enabled, they have immediate access to Agentforce Builder. For anyone else to build agents, they need the Manage AI Agents permission (Salesforce Ben).
Why it matters more at a bank or lender
For most companies, this is a configuration update. At a financial institution it's something else: a generative AI capability became available on top of the customer base without going through the risk, security or compliance committee that normally signs off on any new tool.
That doesn't mean there's a problem. It means a decision that used to be implicit ("it's not on") now has to be made explicitly and documented. We see it on four fronts:
- Who can build. If Manage AI Agents is over-assigned, any admin can put together an agent that reads customer data.
- What data an agent sees. An agent works with the permissions it's given. In a financial org, that can include personal data, balances, loan applications or complaints.
- Traceability. Internal audit will ask what each agent did, with which data and who approved it. It's better to have the answer before the question.
- Spend. Credit consumption is variable. Without an owner and a monthly review, the cost shows up later.
This is our reading, not a requirement published by any particular regulator. Specific obligations depend on each country and each institution, so review them with your compliance team.
The controls that remain
With the Setup option on its way out, these are the controls the sources describe (Salesforce Ben, CETDIGIT):
| Control | What it does | When to use it |
|---|---|---|
| Einstein setting | Turning it off disables the Agentforce platform | If the institution has decided not to use generative AI for now |
| Manage AI Agents permission | Controls who can create and edit agents | Always: give it only to people who build agents |
| Per-agent activation | Turns each agent on or off individually in Agentforce Builder | For controlled pilots and to stop a specific agent |
| Credit consumption review | Shows how much active agents are spending | From the first agent in production, every month |
- Einstein setting
What it does: Turning it off disables the Agentforce platform
When to use it: If the institution has decided not to use generative AI for now
- Manage AI Agents permission
What it does: Controls who can create and edit agents
When to use it: Always: give it only to people who build agents
- Per-agent activation
What it does: Turns each agent on or off individually in Agentforce Builder
When to use it: For controlled pilots and to stop a specific agent
- Credit consumption review
What it does: Shows how much active agents are spending
When to use it: From the first agent in production, every month
Before turning Einstein off, check which other features depend on that setting in your org: some teams may already be using it.
The date you can't miss: December 1
Winter '27 brings another change that can hit banking hard. Starting December 1, 2026, every user who authenticates with SOAP API login() needs the Use Any API Auth permission. Without it, they get an error (Salesforce Developers).
At many financial institutions, Salesforce connects to the core banking system, the loan origination platform or collections tools through integrations that have been running for years. If one of them uses SOAP login() and its user lacks the permission, it stops syncing that day without warning.
On top of that, SOAP login() in API versions 31.0 through 64.0 retires in Summer '27, and Salesforce recommends moving those integrations to OAuth (Salesforce Developers). We list every date in our article on Salesforce Winter '27.
A six-step checklist
- Confirm your org's status. Check in Setup whether Agentforce already shows as on, and check Salesforce Trust for your instance's maintenance window.
- Review who has Manage AI Agents. Keep it to the people who build agents and document why each of them has it.
- Inventory integrations that use SOAP
login(). Test the Use Any API Auth release update in a sandbox before December 1, starting with integration users. - Decide and document the institution's position. Use, pilot or turn off Einstein for now: all three are valid if the right people approve them.
- Assign an owner and a spending cap. One person accountable for agents and a monthly review of credit spend.
- Pick a narrow first use case. Limited data, a named owner and a way to measure whether it works.
A first use case with an owner
If the decision is to move ahead, in our experience the best first agent at a financial institution is internal or low-risk, works with information the customer already knows and has a person reviewing it. Some reasonable examples:
- Application status. Answering what stage a loan or account opening is in, without making decisions.
- Meeting prep for advisors. A customer summary before a call, reviewed by the advisor.
- Pre-due-date reminders. Payment reminders that hand off to a person at the first objection.
For institutions on Financial Services Cloud, Salesforce also offers a dedicated product: Agentforce for Financial Services (Salesforce). Our Financial Services Cloud guide for banks explains how it fits in Latin America, and our Agentforce implementation page shows how we build a first agent.
Do you know what’s been turned on in your org?
Zarasa is a Salesforce Partner, accredited by Salesforce in Agentforce, Agentforce Sales and Agentforce Service. We review your org’s permissions, integrations and consumption in a free 45-minute assessment.
Let’s talk →Frequently asked questions
Does Agentforce turn on by itself in my Salesforce org?
Yes, if your org is eligible. Since Winter '27, Salesforce turns it on by default in orgs running Lightning Experience on Foundations in the Enterprise, Performance, Unlimited and Developer editions, and in orgs with Agentforce 1 Edition. For existing orgs, the rollout started arriving in waves in September 2026.
Does having Agentforce turned on cost money?
According to Salesforce, the default enablement has no additional cost and doesn't change billing. Consumption starts when an active agent performs actions, which are paid for with credits.
Can Agentforce be turned off?
Salesforce is removing the dedicated Setup option. What remains is turning off the Einstein setting, which disables the platform; restricting the Manage AI Agents permission; and activating or deactivating each agent in Agentforce Builder.
What happens on December 1, 2026?
From that date, users who connect with SOAP API login() need the Use Any API Auth permission. Without it they get an error, and the integrations that rely on them stop syncing.
What should a bank or lender review first?
Who holds the Manage AI Agents permission, which integrations use SOAP login() and what the institution's approved position on generative AI is. Once that's settled, you can pick a first use case with an owner and a spending cap.
