Hands On With Agentforce Coworker: Your CRM Search Bar Just Became an AI Agent


Salesforce put an AI teammate directly inside Global Search. Not a separate agent you build and configure, the search bar you already open every day, now reasoning across your records instead of just matching keywords. I enabled Agentforce Coworker from scratch and put it through a real hands on test.

The entitlement check nobody mentions first

Before touching Setup, there's one thing worth knowing. Agentforce Coworker needs Data 360 actually activated in your org, not just provisioned, plus the correct edition and licensing, either seat based through Agentforce 1 Edition or Agentforce for Sales, Service, and Industries licenses, or usage based through Flex Credits.

A plain free Developer Edition won't clear thisFor this build I used a Trailhead Playground specifically provisioned for Agentforce Coworker, which comes with Data 360 already activated and a sample company already loaded. Starting from the official Trailhead module avoids fighting a licensing wall before you've even opened Setup.

Turning it on

The setup itself is short once you know the order.

StepWhere
Assign the Agentforce Coworker Admin permission set to your userSetup > Users
Turn the Agentforce Coworker toggle On, confirm the disclaimerSetup > Get Started with Agentforce Coworker
Grant your own user accessManage Users > Manage Access to Agentforce Coworker

That's the entire setup. Permission set, toggle, grant access. Provisioning the feature itself takes a minute or two once the toggle is flipped.

Search that actually reasons

Open the App Launcher, search Sales Console, and open it. Next to the global search bar there's now a button called Ask.

Ask it something like "give me a summary of open opportunities for this account" and what comes back isn't a list of keyword matched records. It's a reasoned answer pulled from the actual Opportunity records, stage, amount, close date, synthesized into a summary, with inline citations you can click straight through to the record behind each claim.

Multi turn context, no repeating yourselfAsk a follow up like "show me their open cases as well" without repeating the account name. Coworker holds context from the first question into the second. Ask it to reformat as a table and it does that on the spot too.

Orchestrating a separate agent

Coworker isn't only answering questions itself, it's also a front door to any other specialized agent already active in your org. I had a small, separate agent built ahead of time, one job only, adding delivery instructions to a verified customer's Contact record. Built and activated entirely on its own, Coworker had nothing to do with creating it.

Back in the same Coworker conversation, asking it to route the request, "ask one of the agents to add delivery instructions for this customer," Coworker recognized the request matched a job another agent could handle, routed it there, and brought the confirmation straight back into the same conversation. Never left the search bar, never had to know which agent could handle it.

The guardrail nobody mentions

Here's the part worth knowing before rolling this out to a team. Asking Coworker directly to update a Contact's phone number gets a flat refusal, "Unfortunately, I'm not able to directly edit or update records in Salesforce from here."

Off by default, on every objectBy default, Agentforce Coworker cannot touch your data at all, not an Account, not a Contact, nothing. That's a deliberate guardrail, not a limitation to work around. Turning it on is one toggle, Setup > Get Started with Agentforce Coworker > Manage Data > Add Existing Data, Salesforce row, Agent Actions to On. It allows up to fifteen records updated in a single command, and if a team should never have this capability, leaving that toggle off is exactly how you enforce it.

With it on, the same update request gets proposed rather than refused, and Coworker asks for a Confirm before doing anything. One more detail worth knowing, permissions are checked at the moment of confirmation, not before, so insufficient access still blocks the update even after Confirm is clicked.

Key takeaways
  • Data 360 must be genuinely activated, not just provisioned, before Coworker will turn on
  • Search and Ask returns reasoned answers with inline citations, not keyword matches
  • Context carries across a conversation, no need to repeat the account name
  • Coworker can orchestrate a request to a completely separate, already built agent
  • Record updates are off by default and require an explicit Agent Actions toggle
  • Permissions are checked at the moment of confirmation, not before it
Agentforce Data 360 Agentforce Coworker

Setup steps and entitlement details verified against the official Trailhead module and the Agentforce Developer Guide.

Checkout Complete Video on YouTube

 If you have any question please leave a comment below.

If you would like to add something to this post please leave a comment below.
Share this blog with your friends if you find it helpful!

Post a Comment

0 Comments