Salesforce was 'customer zero' for the rebuilt Slackbot; an employee agent that finds company context, drafts work, and connects Slack context with Salesforce data; now also an external product.
Employee request to a drafted briefing or shared Canvas in Slack.
Prepare a customer-meeting briefing
Pulls recent discussions, relevant documents, and customer history into a single, clear briefing before a critical customer conversation[Source 1]
Turn scattered analysis into a shared Canvas
In a documented demonstration, it correlated pilot feedback with an uploaded dashboard image, queried Salesforce for candidate accounts, and wrote the plan into a shared Canvas[Source 3]
Where people stay involved
People review, approve, or guide the agent’s work here.
Reported: Drafts are refined through conversation with the requesting employee; no approval gate beyond that user is documented.
Anthropic Claude (named via VentureBeat); additional providers planned within the year, with Google Gemini named and OpenAI possible[Source 3]
Harness
An employee agent intended as a front door to other agents[Source 1]
Sandbox
Not reportedThe interview describes an LLM with a robust search layer; no execution or isolation boundary is documented.
Tool access
Finds company context, drafts work, manages meetings, connects Slack context with Salesforce data[Source 1]Meeting booking was announced as arriving a few weeks after general availability; calendar availability checking shipped at launch.
Knowledge
Slack context joined with Salesforce CRM data[Source 1]
Context management
Not reportedGrounding in Slack context and conversation continuity are described; how context is maintained or organized over time is not.
Credentials
Permission-aware by construction; sees what the employee can see, respects roles and access controls[Source 1]The documented mechanism is data-access authorization bounded by the employee's permissions; credential storage is not described.
The launch post includes an internal-use account alongside the rollout dates; it does not demonstrate that internal-first release is best for other products.
