Agentic productivity for every employee in Slack.

An AI agent is an intelligent system that can handle customer questions all on its own, no humans needed. It uses machine learning and natural language processing to tackle everything from easy questions to tricky problems, and it can even juggle multiple tasks at once. AI agents can learn and get better over time, which is different from traditional AI that needs humans to step in for certain tasks. AI agents can take action on behalf of employees.

Slack data across channels, threads, canvases and conversations compliments and enriches your connected, enterprise data sources, resulting in more effective agent responses and actions.

Every Slack workspace is a wealth of institutional knowledge – it’s the long-term memory bank of your company. For AI to be useful, it needs access to data, not just the data in one app or system, but across all your systems, including the unstructured data in team conversations. This contextual data is the key to high-quality, relevant AI outputs and adjusting to new contextual clues in real time.

Every company’s untapped, unstructured data – including user-generated content, natural language text, audio and video files and more – enhances agent reasoning and decisioning for better relevance. When you add agents to Slack, customers can give them permission to tap into the public conversational data in your organisation's Slack instance, making them precise and powerful. Because your data is never our product, customers control what data agents can access in Slack. We never train any LLMs on customer data; instead, we use advanced Retrieval-Augmented Generation (RAG) techniques on the unstructured conversational data to give agents the context they need, just at the moment they need it, to support their human counterparts.

There are three types of agents and assistants that you can use in Slack:

1. Agentforce (available Jan. 2025 in Slack): Powered by the most relevant conversational data in Slack and your trusted enterprise data, Agentforce suggests and automates actions on your behalf. Build custom agents that bring agentic productivity to employees right in the flow of work. From human resources to IT, services and sales and beyond, Agentforce can take actions in Slack like creating and updating channels, lists and canvases. Agentforce in Slack will be available soon; talk to sales to learn more.

2. Your custom-built AI Assistants brought to the Slack platform: Right from a dedicated UI in Slack, you can embed AI Assistants that you build using purpose-built APIs. You can build these custom AI Assistants to work the same way you do, so you can offload tasks – from tickets to requests and more – and focus on the important work that needs to get done. These APIs are available today in the Slack Developer Centre.

3. Third-party AI Assistants: Talk to third-party AI assistants to get help with drafting content, surfacing market research or retrieving and summarising files. Today we have several out-of-the-box AI apps ready to use and download from the Slack Marketplace, such as Adobe Express and Cohere.

For third-party AI Assistants, Slack provides guidelines and/or reviews on quality and security before these are published to the Slack Marketplace. Companies building third-party AI Assistants for the Slack Marketplace must agree to abide by Slack’s app guidelines, which include prohibitions on using customer Slack data to train LLMs, and require companies to share important security and compliance details about their apps. Additionally, Slack owners and admins can control which agent apps can be added to their workspaces.

Salesforce’s Einstein Trust Layer was built to help you safely deploy AI across your business. For every Agentforce interaction, the Einstein Trust Layer is working behind the scenes to protect the privacy and security of your data and to promote the safety and accuracy of every response.

• Customer data: we never use customer data to train LLMs.
• Agent guardrails: Customers are in control of what data agents can access in Slack, and create guardrails that keep agents on task and prevent unwanted behaviour.
• Secure LLM gateways: Use foundational LLMs safely by preventing third-party LLMs from retaining or monitoring your customer and enterprise data.
• Zero data retention: After the LLM generates a response and sends it back to Agentforce, it forgets all of that data thanks to our zero data retention agreements, because your data is not our product. Our data masking techniques prevent the LLM from seeing or processing sensitive data, like the names of companies and people.
• AI monitoring: Agentforce will constantly monitor, analyse and detect harmful language embedded in responses. We’re running toxicity scoring models in the background to identify potentially toxic or harmful content.