Slack Introduces AI-Powered Search and Summarization Features for Enhanced User Experience

As an enterprise communications platform, Slack has evolved into a crucial repository for institutional knowledge. However, retrieving that information using traditional search tools has proven difficult. Today, Slack unveiled new features designed to enhance the accessibility of this data, including an AI-powered search tool and the ability to summarize information within channels.

Noah Weiss, Slack's Chief Product Officer, explains that the platform accumulates corporate information informally and in an unstructured manner. The challenge lies in effectively surfacing this wealth of knowledge. “The key takeaway here is that the latest generative AI capabilities enable us to unveil new insights and intelligence from the analysis that's been building up on our platform for years,” Weiss shared.

In May, Slack announced plans to integrate generative AI into its offerings during the Salesforce World Tour in New York City. This initiative was a broad introduction of SlackGPT, a specialized generative AI tailored for content within the Slack environment.

Today’s announcement focuses on implementing these capabilities in concrete ways. Weiss emphasizes that summarizing channel content allows employees to quickly catch up after time away or avoid sifting through lengthy discussions. With the new channel summaries, users can request a synopsis, and Slack's AI generates an overview of the discussed topics, complete with references that illustrate how the summary was crafted—an important aspect of this feature's design, according to Weiss.

“You can explore any section in detail, providing full context. Our aim was to prioritize transparency, build trust, and ensure that users can see our processes and delve deeper if they wish to,” he explained.

Additionally, users are encouraged to ask questions in a conversational tone, similar to interacting with ChatGPT. However, Slack's AI focuses on internal content rather than general internet data. For example, a user could query, “What is Project Gizmo?” and the AI would respond with an answer, also sourced, so users can verify the information's reliability.

Each response includes a quality assessment feature, allowing users to rate the answer as good, bad, or neutral. This feedback helps the AI model improve its accuracy, while the engineering team can monitor the system's performance.

Weiss refrained from discussing the specifics of the underlying AI model but mentioned it comprises a mix of large language models. “We discovered that each model functions differently in terms of speed and quality. We dedicated significant effort to fine-tuning these models for the specific data we have in Slack, as well as enhancing our prompt engineering techniques.”

Slack AI's search and summarization features are available as an add-on for enterprise plans, incurring an additional cost beyond the standard license fee. While pricing details remain undisclosed, the features are currently accessible in the U.S. and U.K., exclusively in English, with plans for additional languages coming soon.