The earlier version of this guide said Agentforce Coworker had no general availability date. Salesforce's announcement on 4 November 2026 changes that plan, though unevenly. Several channels went GA this month, the core setup pages dropped their Beta labels without a matching GA announcement, and a $50 per-user, per-month SKU went on sale. The decisions in front of you now are which licence to run Coworker on and the order in which to open its channels.
Status in November 2026 and where each channel stands
Coworker, previously called Ask Agentforce, has more than 185,000 users and 21x usage growth over three months, according to the newsroom post. That post applies "generally available" to a specific list of items, never to Coworker as a whole, and it carries the usual forward-looking disclaimer about unreleased features.
| Item | Status | Source |
|---|---|---|
| Coworker in Microsoft Teams | GA, November 2026 | Salesforce Newsroom, 4 Nov 2026 |
| Coworker app for iOS | GA, November 2026 | Salesforce Newsroom, 4 Nov 2026 |
| Desktop app for Mac and Windows | GA, November 2026 | Salesforce Newsroom, 4 Nov 2026 |
| Coworker MCP server | GA, November 2026 | Salesforce Newsroom, 4 Nov 2026 |
| Free trial for Flex Credits customers | GA, November 2026 | Salesforce Newsroom, 4 Nov 2026 |
| Google Workspace and Microsoft 365 connectivity | Available now | Salesforce Newsroom, 4 Nov 2026 |
| Conversation History and Skills | Available now | Salesforce Newsroom, 4 Nov 2026 |
| $50 per user per month SKU | Available now; pricing subject to change, varies by region | Salesforce Newsroom, 4 Nov 2026 |
| Daily Digest and Scheduled tasks | Expected December 2026 | Salesforce Newsroom, 4 Nov 2026 |
| Coworker in Field Service, Consumer Goods Cloud and Life Sciences Cloud mobile apps | Expected December 2026 | Salesforce Newsroom, 4 Nov 2026 |
| Core Coworker product | No GA announcement; Turn On and user access pages fetched 6 Nov 2026 carry no Beta label | developer.salesforce.com setup guide |
Search engine listings for those same setup pages still show "(Beta)" titles, and the earlier guide carried the Beta Services Terms notice. I treat the November channel items as GA and the core product as unconfirmed. Check the labels in your own Setup, read your order form, and ask your account executive to confirm in writing before you assume beta terms no longer apply. Under the beta terms the earlier docs carried, credits consumed by non-GA services carry no refund or credit rights.
Rollout order
This is the order I would run. Pilot in the Salesforce browser experience with people whose identity mappings you have checked by hand (covered below). Once the answers look right and you understand the credit draw, open Teams or the desktop app to that group, whichever they already live in. iOS comes last, because a phone is where people are least likely to question a wrong answer. Daily Digest, Scheduled tasks and the industry mobile apps wait for December, when their setup docs should exist. Every new surface is another sign-in, and the setup docs don't yet describe how each channel handles identity, so fix mappings before adding channels.
Choosing a licence: per-user SKU, Flex Credits or seat-based
Every path needs Enterprise, Unlimited or Agentforce 1 edition, plus "Data Cloud or Data 360 (provisioned and activated)". The Coworker setup home page says it is not supported in Government Cloud Plus.
| Path | What the org holds | How cost behaves | Extra setup step | When I would pick it |
|---|---|---|---|---|
| Per-user SKU | The new $50 per user per month option | Flat per named user | Probably the Unmetered User tab (inference, see below) | Daily users with predictable usage |
| Usage-based | Flex Credits, Einstein Requests or Data Services Credits entitlements | Draws down with use | Agentforce Coworker tab only | Pilots, occasional users, the free trial |
| Seat-based via Agentforce bundles | Agentforce 1 Edition, or Agentforce for Sales, Service or Industries licence | Covered by the bundle you already own | Unmetered User Based AI setup first (see the assignment table) | You already pay for the bundle |
| Seat-based Coworker licence | Agentforce Coworker for Enterprise License or for Unlimited Edition License | Per seat | Also assign the Unmetered User tab | Same case as the per-user SKU, once your AE confirms they match |
The docs do not say that the Coworker for Enterprise and Unlimited Edition licences are the $50 SKU. That is a reasonable reading, and I would confirm it with the AE before signing. Flex Credits customers get a free trial, but Salesforce has not published its length or limits, so ask for both.
Two costs sit outside any per-user price. Unmetered prompts do not cover Data 360 ingestion, processing or indexing, which draw on Flex Credits or Data Services Credits, and the docs say building a search index for Data 360 objects "can consume credits". Take a Digital Wallet baseline before the pilot so you can attribute the change to Coworker.
For the per-user decision, compare seats bought with people who hold access:
SELECT MasterLabel, TotalLicenses, UsedLicenses
FROM PermissionSetLicense
WHERE MasterLabel LIKE '%Coworker%'
OR MasterLabel LIKE '%Unmetered%'
SELECT PermissionSet.Label, COUNT_DISTINCT(AssigneeId) activeUsers
FROM PermissionSetAssignment
WHERE (PermissionSet.Label LIKE '%Coworker%'
OR PermissionSet.Label LIKE '%Unmetered%')
AND Assignee.IsActive = true
GROUP BY PermissionSet.Label
This query counts assignments, so treat Digital Wallet consumption during the pilot as the usage signal. If the pilot shows most of a team using Coworker every day, the flat SKU is easier to forecast. If half the permissioned users rarely open it, I would stay usage-based and remove the idle assignments.
Turn On, permissions and user access
Turn On sequence
The admin requirement is now "System Admin Profile and Agentforce Coworker Admin Permission Set". Earlier docs said Salesforce Admin Role, so check again who on your team can actually run setup.
- Confirm Data 360 is provisioned and activated; provisioned alone is not enough.
- In Setup, enter Agentforce Coworker in Quick Find.
- Open Get Started with Agentforce Coworker, select Turn On, then Confirm.
- Wait. The docs say setup can take a few minutes.
- Check that Einstein switched on. It should happen automatically, and I verify it every time.
- If an automated step fails, a warning with instructions appears on the setup page. Clear it before assigning anyone.
Assigning and removing access by licence type
Assignment starts at Setup > Get Started with Agentforce Coworker > Manage Users > Manage. On the Agentforce Coworker tab, select users and Assign. What happens next depends on the licence.
| Licence type | Assign | Remove (Setup > Users > select user) |
|---|---|---|
| Coworker for Enterprise or Unlimited Edition | Agentforce Coworker tab, then the same users on the Agentforce Coworker Unmetered User tab, then Done | Agentforce Coworker User and Agentforce Coworker Unmetered User |
| Agentforce 1, or Agentforce for Sales, Service or Industries | First assign the Unmetered User Based AI permission set license, create an Unmetered User Based AI permission set, and assign it (the current docs link this as "Configure User Permissions for Unmetered Licenses"), then the Agentforce Coworker tab, then Done | Unmetered User-Based AI and Agentforce Coworker User |
| Usage-based | Agentforce Coworker tab, then Done | Agentforce Coworker User |
Auditing old and new names
The earlier docs named the end-user assignment the Access_Ai_Search permission set group with an AI Search permission set licence, and removing access meant taking off both. Orgs set up during the beta may still carry those next to the new sets, and a removal that drops only the group leaves the licence assigned to someone who can no longer use Coworker. Audit both generations:
SELECT Assignee.Username, Assignee.IsActive,
PermissionSet.Label, PermissionSetGroup.DeveloperName
FROM PermissionSetAssignment
WHERE PermissionSet.Label LIKE '%Coworker%'
OR PermissionSet.Label LIKE '%Unmetered%'
OR PermissionSetGroup.DeveloperName = 'Access_Ai_Search'
ORDER BY Assignee.Username
SELECT PermissionSetLicense.MasterLabel, Assignee.Username, Assignee.IsActive
FROM PermissionSetLicenseAssign
WHERE PermissionSetLicense.MasterLabel LIKE '%Coworker%'
OR PermissionSetLicense.MasterLabel LIKE '%AI Search%'
OR PermissionSetLicense.MasterLabel LIKE '%Unmetered%'
Rows where Assignee.IsActive is false are seats you can reclaim.
Data access, identity mapping, record writes and companion orgs
The newsroom says Coworker runs on each employee's permissions, sharing rules and configured access controls. The detail below was last confirmed in the October 2026 revision of the setup docs, so recheck it against your org.
Access policies
Connecting Google Drive or SharePoint applies default data access policies automatically. Slack results follow Slack's own visibility. For object-level and record-level governance in Data 360 you need the Data 360 Admin or Data 360 Architect role, and the Data Governance tab needs Data 360 Architect. Go to App Launcher > Data 360 > Data Governance tab > Policies.
At object level, create two policies, on the Search Query DMO and the Record View DMO, granting access to every user who holds a Coworker permission set or licence. At record level, create four policies, on the Record View DLO, Record View DMO, Search Query DLO and Search Query DMO, denying records where the Individual field does not match the current user ID. Any new DMO you build from the Record View DLO or Search Query DLO needs its own policy, and that step is easy to forget six months after the pilot.
Identity mapping
Turn On creates an Identity Resolution ruleset that links each person's Salesforce CRM, Slack and Google Drive identities, matching on email or federated ID by default. A profile from one source inherits permissions from its linked source, so a wrong match is a permissions problem. To check, enter User Mapping in Setup Quick Find, select a user, and read the name, username and email the match used. Check by hand anyone whose Slack or Google email differs from their Salesforce email.
Record writes
Writes stay off until an admin opens Get Started with Agentforce Coworker, finds the Salesforce CRM row in the Add Existing Data table and switches on Agent Actions. Users then confirm each proposed change in Ask mode. In the October docs:
- Object and field permissions are checked only at confirmation, so a user can be shown a change they cannot make, and it then fails.
- A single command can change at most 15 records.
- The toggle applies to the whole source, with no scoping by user, object or field.
- Only field-level required is enforced, so fields marked required on the page layout can be saved empty.
- Proposed changes may skip field dependencies and validation rules, so a save can fail after the user confirms.
- No separate Coworker audit log was documented. Standard field history records the changes as the user's own edits.
The newsroom now mentions "approval and audit capabilities available for supported operations". Test what your org actually records before relying on it. I would leave Agent Actions off for the first wave and turn it on for a second wave that works mostly on objects with simple validation.
Companion org (Data Cloud One)
Set up both orgs, create a data space and share it with the companion. If Coworker is already on in the home org, share that org's data space or choose a different home org. Turn Coworker on in the home org. In the companion, check that Connection Status is Active, Companion Sync Status is Succeeded and the data space Initialization Status is Active, add a Data Cloud Governance policy on the shared data space (Allow All Objects suits most orgs, and a custom policy gives tighter control), then turn on Coworker there.
Coworker MCP server or Hosted MCP Servers
The Coworker MCP server is listed as GA for connectivity, but the announcement does not say whether it exposes Coworker to external MCP clients or lets Coworker call external MCP servers. If you search for its docs, watch for Coworker AI (coworker.ai), an unrelated company that sells a product also called Coworker MCP.
| Axis | Coworker MCP server | Hosted MCP Servers |
|---|---|---|
| Who runs it | Salesforce | Salesforce, which also handles sign-in |
| Direction | Not yet confirmed in setup docs | External assistant calls into Salesforce |
| What it reaches | Not yet confirmed | Record lookups, Flows and Apex actions |
| Permissions | Coworker uses employee permissions and sharing (newsroom) | Salesforce permissions enforced |
| Scope | Not yet confirmed | Salesforce only; other tools need their own integrations |
| Metering | Not yet confirmed | Headless Platform Interactions against Flex Credits; multiplier unpublished, 30 days' notice committed |
Agentforce Coworker or a Hosted MCP Servers build
MCP is an open standard for connecting AI assistants and agents to other systems. With Hosted MCP Servers, Salesforce runs the server, handles sign-in and enforces permissions, and an assistant outside Salesforce can look up records and run your Flows and Apex actions. Most teams weighing MCP are choosing between Coworker itself and that kind of build.
| Concern | Agentforce Coworker | Hosted MCP Servers |
|---|---|---|
| Who runs it | Salesforce, inside your org | Salesforce runs the server, handles sign-in and enforces permissions |
| Where users work | The Coworker experience in Salesforce, Microsoft Teams, the iOS app and the Mac and Windows desktop app | An external assistant of your choice |
| Data reach | Salesforce CRM, Data 360, Slack, Google Drive and SharePoint sources, plus Google Workspace and Microsoft 365 connectivity | Salesforce only; Google, Slack and other tools need their own integrations |
| Actions | Record writes through Agent Actions, plus delegation to your other Agentforce agents | Record lookups, Flows and Apex actions |
| What you build | Configuration and policies | Logic, context assembly and UX |
I would choose MCP when users already spend their day in another assistant, when you need your own orchestration or UX, or when the org isn't eligible for Coworker. Running both is fine. On the MCP side, one user question can trigger several tool calls, so model cost per question; our Flex Credits agentic API post covers the metering plan.
Whichever path an agent takes, the Apex actions you expose decide what it can read. A common failure is an Apex action written in system mode for a nightly integration and later exposed to an agent. Users can then read fields in chat, such as a credit balance, that their profile hides on the record page. Write agent-facing actions in user mode from the start:
public with sharing class OpenCasesByPriority {
public class Request {
@InvocableVariable(required=true label='Account Id')
public Id accountId;
}
public class Result {
@InvocableVariable(label='Summary')
public String summary;
}
@InvocableMethod(label='Count Open Cases by Priority'
description='Counts open cases on an account that the running user can see')
public static List<Result> countOpenCases(List<Request> requests) {
Set<Id> accountIds = new Set<Id>();
for (Request req : requests) {
accountIds.add(req.accountId);
}
Map<Id, List<String>> linesByAccount = new Map<Id, List<String>>();
String failure;
try {
for (AggregateResult row : [
SELECT AccountId acct, Priority pri, COUNT(Id) total
FROM Case
WHERE AccountId IN :accountIds AND IsClosed = false
WITH USER_MODE
GROUP BY AccountId, Priority
]) {
Id acct = (Id) row.get('acct');
String pri = (String) row.get('pri');
if (!linesByAccount.containsKey(acct)) {
linesByAccount.put(acct, new List<String>());
}
linesByAccount.get(acct).add(
(pri == null ? 'No priority' : pri) + ': ' + (Integer) row.get('total'));
}
} catch (QueryException e) {
failure = 'You do not have access to case data for this account.';
}
List<Result> results = new List<Result>();
for (Request req : requests) {
Result res = new Result();
if (failure != null) {
res.summary = failure;
} else if (linesByAccount.containsKey(req.accountId)) {
res.summary = String.join(linesByAccount.get(req.accountId), ', ');
} else {
res.summary = 'No open cases you can see on this account.';
}
results.add(res);
}
return results;
}
}
The with sharing keyword limits rows to the user's sharing. WITH USER_MODE adds object and field security and throws a QueryException when the user lacks access, which the action turns into a plain message for the agent. The method takes a list because invocable actions are always called in bulk form, even for a single request.
What to watch for
- The core product has no GA announcement. Get beta terms resolved in writing before a wide rollout.
- Per-user pricing is subject to change and regional availability, and the docs do not tie the Coworker Enterprise and Unlimited licences to the $50 SKU.
- Data 360 ingestion, processing and indexing draw credits on every licence path.
- Beta-era orgs may hold the Access_Ai_Search group and AI Search licence alongside the new permission sets.
- Users whose email differs between Salesforce, Slack and Google Drive can be mapped wrongly and inherit the wrong permissions.
- Agent Actions checks permissions only at confirmation and may skip validation rules and layout-required fields.
- Daily Digest, Scheduled tasks and the industry mobile apps are December items with no setup docs yet.
- The Coworker MCP server's direction, setup and metering are not yet confirmed.
Leave a Comment