Slack

Slack is a workplace collaboration platform built around channel-based messaging, connected software, and searchable organizational knowledge. Salesforce acquired Slack Technologies, Inc. in July 2021. Its later AI products use workplace conversations and files as context: Slack AI introduced search answers and summaries in 2024, and the reimagined Slackbot expanded into personal assistance and actions in 2026.[9][10][3][5]

Slack began rolling out Slack AI on February 14, 2024, initially in English as a paid add-on for Enterprise plans. Its first capabilities were personalized search answers, channel recaps, and thread summaries. Search answers cited relevant messages; summaries also cited their sources so users could inspect the underlying conversation. Channel recaps could cover unread messages, the previous seven days, or a custom date range—for example, helping someone joining a sales deal catch up on the account channel.[3]

Slack engineers Kelly Moran, Curtis Allen, and Alex Benjamin explained the architecture in an April 2024 account, updated in July 2025. The team chose off-the-shelf models rather than training or fine-tuning generative models on customer data. It used retrieval augmented generation (RAG): each request supplied the task’s context and instructions, such as channel messages and directions for summarizing them. AWS acted as a broker for hosting closed-source models in an escrow VPC within Slack’s trust boundary, excluding model-provider access to customer data.[4]

That design made context capacity and latency consequential constraints: summarizing a busy channel required a large context window, while sending more messages slowed processing. The team evaluated models for search and summarization, then improved results through prompt tuning and chaining traditional machine-learning models with generative models.[4]

Slack AI fetched content using the requesting user’s access-control lists and existing channel and search retrieval libraries. It therefore operated on content the user could already see. The engineering account also described point-in-time search answers and conversation summaries that were not stored on disk where possible. When data-loss-protection controls tombstoned a source message, summaries derived from it were invalidated.[4]

At launch, Slack reported that an internal pilot analysis involving customers such as SpotOn, Uber, and Anthropic estimated potential savings averaging 97 minutes per user each week. It also reported more than 13,400 customer-built AI custom internal Slack apps. Partner integrations served distinct purposes: PagerDuty Copilot supported incident response, Notion supplied AI summaries when links were shared, and Perplexity piped topic updates into Slack.[3]

Salesforce announced general availability of the new Slackbot on January 13, 2026, for Business+ and Enterprise+ customers through a phased rollout scheduled across January and February. Built directly into Slack, the personal agent could find answers, draft content, organize work, and schedule meetings. It combined permitted Slack conversations and files with Salesforce customer data to prepare briefings and identify next steps. Broader coordination with Agentforce and third-party agents was described as future functionality.[5]

Slack also became an interface for agents developed by other companies. A June 24, 2026 Latent.Space article reproduced Anthropic’s June 23 announcement of Claude Tag: Claude would join Slack as a team member with access to selected channels and tools, allowing people to tag it into tasks and continue other work. Claude Tag was Anthropic’s product, extending agent participation into Slack’s team conversations.[2]

The interface’s potential also attracted criticism. In a February 14, 2026 editorial advocating an OpenAI competitor, swyx criticized Slack AI’s discoverability and personalization and argued that huddles could offer better real-time multimodal AI features. He singled out cross-company collaboration for praise: “Slack Connect is great though, definitely just clone that.”[1]

These AI products followed an earlier period of financing, customer growth, and acquisition. Slack had already introduced machine-learning-powered features in 2016 and used traditional ML to rank search results; its later generative systems built on existing search and access-control infrastructure.[4][8][10]

On August 21, 2018, Slack announced a $427 million Series H equity round led by Dragoneer Investment Group and General Atlantic, at a post-money valuation above $7.1 billion. The announced investment was additional to $841 million previously raised. At that date, Slack reported more than 8 million daily active users and more than 70,000 paid teams.[8]

For the fiscal year ended January 31, 2021, Slack reported revenue of $902.6 million, up 43% year over year, and over 156,000 Paid Customers, up 42%. Slack Connect adoption reached over 74,000 Paid Customers and over 660,000 connected endpoints. These were separate metrics: annual revenue, customers, and endpoints did not represent individual active users.[9]

Salesforce completed its acquisition of Slack on July 21, 2021. At closing, it said Slack would retain its brand and continue under CEO and co-founder Stewart Butterfield. Salesforce positioned the combination around connecting employees, customers, partners, and applications with Customer 360—a business-software context that preceded Slack’s later AI search and personal-agent products.[10][3][5]