Agentforce Coworker: What Salesforce Admins Should Check

Agentforce Coworker is growing fast and may already be on in your org. See what it does, what it costs, and five checks admins should run now.

ConvoPro Team

Salesforce AI Workflow Advisors

Featured

agentforce coworker

Agentforce Coworker: What Salesforce Admins Should Check Before Users Delegate Work

Agentforce Coworker is Salesforce’s AI teammate for employees. Salesforce describes it as a way to delegate business work using trusted Salesforce context: Coworker plans the steps, coordinates the right Agentforce agents and workflows, and brings back results and recommended actions for review.

It is also moving quickly. In its November 4, 2026 Agentforce Coworker announcement, Salesforce reported more than 185,000 users and 21x usage growth in three months, introduced a $50 per user, per month option, and listed Microsoft Teams, desktop, and iOS apps as generally available in November.

For many orgs, the more pressing detail is older. Salesforce began automatically enabling Coworker for customers with unmetered user-based AI seats in August, so some admins are now governing a tool they never chose to roll out. This guide covers what Coworker does today, what the public documentation does and does not say, five checks to run now, and where Coworker fits next to Salesforce Flow and other workflow tools.

What Agentforce Coworker does today

Salesforce’s materials describe Coworker in two layers, and it helps to read them separately.

Inside Salesforce: Ask from Global Search

The Agentforce Coworker automatic enablement FAQ describes an Ask button next to the Global Search bar. In Ask Mode, users get AI answers drawn from Accounts, Contacts, and Knowledge. In Search Mode, Agentforce responses appear alongside standard search results, and orgs can optionally connect Slack and Data 360.

Across the tools employees already use

The November announcement describes a broader role. Google Workspace and Microsoft 365 connectivity is available now, so Coworker can combine email, calendars, and files with Salesforce data. A Teams app, Mac and Windows desktop apps, an iOS app, and a Coworker MCP server are listed as generally available in November 2026, with a daily digest and scheduled tasks expected in December.

The practical reading for admins is that Coworker is expanding from answering questions in search to coordinating multistep work across apps. Govern it as a capability whose scope will keep growing, not as a search enhancement you can set and forget. Salesforce is widening its agent lineup in the same cycle, and our look at Agentforce job-ready agents for lean Salesforce teams covers that side of the announcements.

Why Coworker is now an admin decision

According to Salesforce’s help documentation, automatic enablement began on a rolling basis on August 10, 2026, for customers with unmetered user-based AI seats. Admins are not required to take any action. Salesforce sends an email before activation, and an in-product banner confirms when Coworker is active.

Several orgs are excluded from automatic enablement, including orgs that chose not to enable generative AI, multi-org or multi-dataspace topologies, and HIPAA, health and life sciences, and government orgs. Orgs that rely only on usage-based seats without unmetered entitlements are also excluded. Salesforce states that Coworker is not currently covered by its Business Associate Addendum, which matters for any team handling protected health information.

None of this makes Coworker risky by default. Salesforce says Coworker is designed to operate within its established security and governance framework, using applicable employee permissions, sharing rules, and configured access controls.

The catch is that an assistant which respects your permissions will faithfully reflect them. If sharing is broader than it should be, AI answers will be broader too. The NIST AI Risk Management Framework makes the same point in general terms: governance starts with knowing what a system can reach and who is accountable for what it produces.

Five checks to run this month

Check 1: Confirm whether Coworker is already on

Look for the activation banner and open Agentforce Coworker Setup. If your org has not been enabled yet and you want more time, Salesforce documents an opt-out path in Setup under Agentforce Coworker, Opt Out of Auto Activation, as well as a link in the notification email. After enablement, the toggle can be turned on or off at any time.

Also plan for users who join later. Users who gain AI entitlements after your org is enabled see standard search until an admin grants access, which gives you a natural checkpoint for each new team.

Check 2: Audit what users can see before AI summarizes it

Review org-wide defaults, sharing rules, permission sets, field-level security on sensitive fields, and Knowledge article visibility. Pay close attention to records that were technically visible but rarely found through manual search, because a conversational interface makes them easy to surface.

Our Salesforce AI security checklist for admins and CIOs walks through the access questions to answer first.

Check 3: Decide which changes need a human before records move

Salesforce says Coworker brings back results and recommended actions for review, and that approval and audit capabilities are available for supported operations. The public documentation does not list every supported action or which ones require approval, so test write behavior in a sandbox before relying on it.

Write down, object by object, which changes an AI may draft and which require a person to approve. For a starting framework, see where humans should review before Salesforce records change.

Check 4: Model cost before adoption spreads

Salesforce lists the $50 per user, per month option as available now and notes that pricing and packaging are subject to change. It also lists a free trial for Flex Credits customers in November 2026. For orgs with qualifying seats, the help documentation states that searching CRM data and Slack through Coworker does not consume Flex Credits or Data Services Credits, while Data 360 ingestion, processing, and indexing consume credits regardless of billing plan.

Watch volume as well as price. The help documentation sets a limit of 30 LLM calls per minute per user for seat-based licenses, notes that a single question can trigger several calls, and says the limit is not currently technically enforced, although Salesforce reserves the right to limit usage. Confirm contract terms with your account executive, and use our Agentforce cost calculator to frame the math at the workflow level.

Check 5: Review connected apps and regulated data

Google Workspace, Microsoft 365, Slack, Data 360, and the new MCP server each widen what Coworker can draw on. Decide which connections are approved, who owns each one, and how often you will review them.

If you plan to connect external AI clients through MCP, our Salesforce MCP security checklist covers identity, tool permissions, approval, and audit.

Where Coworker fits next to Flow and workflow tools

Coworker is built for employees. It is strongest when the person asking is an authenticated Salesforce user, the data already lives in the org or a connected workspace, and the job is research, summarizing, coordinating, or recommending next steps.

It is not designed around work that starts with people who are not Salesforce users. A customer reporting an equipment fault, a vendor submitting onboarding documents, or a field technician scanning an asset tag all produce messy input that has to be structured into required fields and reviewed before a record should exist.

The table below shows how common options divide that work. Each one complements the others, and Salesforce remains the system of record.


Option

Best fit

Look elsewhere when

Salesforce Flow

Stable, rules-based processes that run entirely inside Salesforce

Inputs are unstructured or arrive from outside Salesforce

ConvoPro

Messy input from email, files, forms, or QR codes that must become a reviewed, structured Salesforce record

You need a general AI assistant for every employee across every app

Agentforce Coworker

Employees delegating research, summaries, and multistep work using Salesforce context

The submitter is not a Salesforce user, or input must be mapped into required fields before any record exists

Custom Agentforce agents

Role-specific or customer-facing agents built, tested, and orchestrated at scale

The team lacks capacity to design, test, and maintain agents

Example: equipment service requests

Consider a mid-market team that services medical equipment in the field. Requests arrive as customer emails with photos of an error screen, phone calls relayed by technicians, and QR codes printed on devices. Today, a coordinator reads each one and re-keys the details into a Case.

Flow alone struggles here because it expects structured inputs, and the coordinator is the structuring step. A governed intake layer like ConvoPro can take on that step: a QR scan opens a guided form with the asset context attached, collects notes and photos, maps the input to Case fields, and routes the proposed record to a reviewer before it is created. That review-before-create pattern keeps bad data out of the org at the point of entry.

Once clean Cases exist, Coworker becomes more useful. A service manager can ask Coworker to summarize open Cases on an account, pull related Knowledge articles, and suggest next steps for review, while Flow continues to handle rules-based routing and escalation. The tools stack, and the quality of intake determines how much value every assistant downstream can deliver.

Your next step

If Coworker is already active in your org, run the five checks above, starting with access. Governance done now is cheaper than cleanup after usage spreads, and our Salesforce agent governance playbook covers ownership and review cadence for every agent in your org.

If you also have a workflow that starts outside Salesforce, runs 20 or more times a week, and ends in re-keyed data, that is the place to prove a governed intake layer. You can start a free ConvoPro Studio trial and work through Salesforce-style workflow examples before connecting your org, or review ConvoPro pricing to scope a pilot.

Frequently asked questions

Is Agentforce Coworker turned on automatically?

For customers with unmetered user-based AI seats, Salesforce began rolling out automatic enablement on August 10, 2026. Some orgs are excluded, including those that opted out of generative AI, multi-org topologies, and HIPAA, health and life sciences, and government orgs. Admins can opt out before activation or turn Coworker off afterward in Agentforce Coworker Setup.

How much does Agentforce Coworker cost?

Salesforce’s November 2026 announcement lists a $50 per user, per month option and a free trial for Flex Credits customers, and states that pricing and packaging are subject to change. Data 360 ingestion, processing, and indexing consume credits separately. Confirm the terms that apply to your contract with your Salesforce account executive.

Does Agentforce Coworker respect Salesforce permissions?

Salesforce states that Coworker respects user access permissions and is designed to operate within existing permissions, sharing rules, and configured access controls. That also means answers will reflect any sharing that is broader than intended, so review access before usage grows.

Can Agentforce Coworker create or update Salesforce records?

Salesforce describes Coworker returning results and recommended actions for review, with approval and audit capabilities for supported operations. The public documentation does not list every supported action, so test write behavior in a sandbox and define which changes require human approval.

How does Agentforce Coworker relate to Salesforce Flow and ConvoPro?

They solve different parts of the same process. Flow automates rules-based steps inside Salesforce, ConvoPro turns messy input into reviewed, structured Salesforce records, and Coworker helps employees delegate work using the data those tools produce. Many teams will use more than one.

Share on social media