Three months ago, Agentforce Coworker — the AI teammate living inside your Salesforce interface — had 100,000 pilot users. As of the November 4 announcement, usage has grown 21 times in three months, past 185,000 users, and Salesforce has stopped treating it as a Lightning feature. Coworker is now expanding into Microsoft Teams, iOS, and Mac and Windows desktops, with Google Workspace and Microsoft 365 connections already live.

Read the product logic carefully, because it is a genuine shift in how enterprise AI gets consumed. The pitch is no longer "here are more agents to manage." It is: you give one teammate a task in plain language; it pulls the context, coordinates the right agents and workflows, and brings back a result for you to review. Salesforce's SVP James Nakashima framed it exactly that way — not more agents for users to babysit, but one Coworker that rallies the right ones.

What shipped, and when

Capability Status
Google Workspace & Microsoft 365 connections (email, calendar, files in tasks) Live now
Conversation History & Skills (reusable best practices) Live now
Per-user subscription — $50 per user per month Live now
Microsoft Teams integration; iOS, Mac & Windows apps; Coworker MCP server GA during November 2026
Daily Digest & scheduled tasks Expected December 2026
Field Service, Consumer Goods & Life Sciences mobile integrations On the roadmap

The example Salesforce demonstrated is the whole story in miniature: a sales rep asks Coworker to investigate an at-risk renewal. It pulls the service history, finds open cases, prepares the follow-up, and coordinates the relevant agents and workflows — then hands the results back for human review. The employee assigns a business task; the governance context, permissions, and Data 360 connections do the rest.

The adoption numbers behind the press release

Two customer proof points came with the announcement, and both are reported figures worth quoting carefully. Adecco Group is rolling Coworker out across 40 countries — their CEO describes it as the single place employees go to view candidate profiles and client context — on top of reported results including 93% faster job-campaign prep and candidates screened by Agentforce being twice as likely to be hired. Fulton Bank has deployed more than 20 Coworker use cases covering roughly 3,000 employees. Whatever you think of vendor-reported numbers, 21x growth in three months is the steepest adoption curve Salesforce has shown for any product in recent memory.

The pricing shift deserves its own sentence. A per-user, $50-per-month SKU turns Coworker from a platform experiment into a line item — which means budget owners will start asking the question you should be ready for: which roles get an AI teammate first, and how do we measure what they got back?

What to learn while everyone else is reading headlines

Role The skill that matters
Administrator Data quality and permissions — Coworker inherits the user's access, so an over-permissioned user now has an over-permissioned teammate. Data hygiene stopped being cosmetic
Consultant Task decomposition — deciding which work belongs to Coworker, which to dedicated agents, and where the human review gate sits
Developer The Coworker MCP server and Skills model — packaging team best practices into reusable capabilities instead of one-off prompts
Career-switcher / student Judgment over tooling — every employer will have an AI teammate; few will have people who know when to check its work

One pattern is worth naming before it becomes obvious to everyone. Coworker's expansion — Teams, desktop, email, calendar, files — means the AI teammate now sits inside the same surfaces where your permissions, your data quality, and your business logic already live. The orgs that struggled with data quality in the chatbot era will struggle visibly with a teammate that reads everything. The orgs that cleaned up their data will look like they hired a miracle worker.

So the question to sit with this quarter is not "should we buy Coworker?" At 185,000 users and a December roadmap, someone in your company will ask it for you. The question is: when they do, will you be the person who already knows what a good Coworker task looks like, where it needs a review gate, and which data it should never touch? That person is not competing for an AI job. They are competing for the job of making AI work — and right now, almost nobody else is preparing for it.