Slack is working on new generative AI features that provide users with automated summaries of conversations and AI assisted writing natively within its collaboration app.

Slack had already discussed plans to connect its popular software to large language models (LLMs) offered by third-parties, announcing an integration with OpenAI's ChatGPT (currently in beta) in March, and Anthropic's Claude, available now. It expanded on those plans Thursday at parent company Salesforce's World Tour event with the unveiling of Slack GPT, which will see generative AI integrated directly into the Slack app in a variety of ways.

The goal is to improve sharing and creation within Slack and tie it in to Salesforce's products more effectively, the company said.

Slack's popularity as a workplace collaboration platform provides it with some important advantages when incorporating generative AI, said Ali Rayl, Slack's vice president of customer experience.

"We think that Slack has a unique advantage when it comes to generative AI," she said. "First and foremost, we hold an enormous amount of institutional knowledge: everything that your entire team has done in channels for the years that you've been using Slack, it's in your archives, it's all at your fingertips. That's an enormous body of knowledge for us to tap into.

"Given the reliance we all have on chat-based collaboration, leveraging generative AI to summarize and automate what we have to do now - labor intensive scrolling around, switching from chat stream to chat stream, uncovering the key people in those endless threads - these types of initial applications of generative are helpful advancements," she said.

One focus for Slack GPT is conversation summaries.  For example, a user facing lots of unread messages when returning to a team channel conversation can click a button for an overview of what's been discussed while they were away, enabling them to quickly catch up without scrolling through numerous posts. It will be possible to do the same with transcriptions of Slack Huddle, the company's lightweight voice meeting tool.

Another feature under development is AI-assisted writing. Many people already use ChatGPT for this purpose, whether it's to suggest an email response or make an existing draft more formal in tone. Rather than copy/pasting this information to and from ChatGPT, Slack will enable these options via a small dropdown menu.

Slack GPT can also create information in Canvas documents. "We envision a world where AI can start populating canvases for you," said Rayl. This would mean generating text in the document, adding profile cards of those involved in a particular project, and including relevant videos and documents.

Slack also sees ways to add LLMs from the likes of OpenAI and Anthropic into its no-code Workflow Builder to create more powerful automations.

Rayl gave an example of an "AI-powered lead response." This could start with a new lead coming into the Salesforce app, which then kicks off a process where the relevant Slack team channel is notified with any background information about the customer and an email is generated in Gmail for review before it's sent. "This is one really quick workflow that anybody can build that really shaves off the amount of time from a salesperson getting a lead to actually reaching out to that customer," she said.

The ability to integrate generative AI with the Workflow Builder is slated to be available this summer.

Another aspect of Slack GPT involves the Einstein GPT and the ability to access a range of business information across the Salesforce Customer 360 software portfolio. "You can use Slack as the conversational interface to say, 'Einstein, give me information about this customer, about this campaign,' and get a good summary of what's going on in your Salesforce products or Salesforce Customer 360 right in Slack," said Rayl.

Slack did not say when the native features or Einstein GPT are expected to launch.

With gathering interest in the business potential for LLMs, many enterprise software vendors - including Box, Zoom, and Atlassian - are looking to incorporate generative AI into their products. In that respect, Slack GPT is a natural direction for the company.

"Given the reliance we all have on chat-based collaboration, leveraging generative AI to summarize and automate what we have to do now - labor-intensive scrolling around, switching from chat stream to chat stream, uncovering the key people in those endless threads -these types of initial applications are helpful advancements," said Mike Gotta, research vice president for collaboration, employee communications, and social software at Gartner.

Along with the benefits of generative AI, organizations using these tools also must contend with potential problems involving  "security, confidentiality, and scaling..., which is challenging at such an early stage and with so many vendors making news in the space," said Gotta.

"There is a certain level of exuberance always expected when new technology emerges," Gotta said. "But if we've learned from the past, it takes some time to ensure you can operationalize the technology in ways that improves employee experience but also handles business complexities, including integration, security and governance."

Data privacy is another issue. Samsung, for example, which recently banned its staffers from accessing ChatGPT after sensitive data was posted to the AI platform. However, Rayl said Slack conversation data will not be made available to external parties, and the company has a "no storage, no training" contract in place with OpenAI and Anthropic for its app integrations. In this way it will offer greater security to businesses where staff want to use LLMs in their jobs, she said.

Finally, there remain ongoing concerns about the accuracy of the information created. Tools such as ChatGPT often present information that's wildly inaccurate in an authoritative manner.

There are ways to address this downside in product design, said Rayl. It's important that users are aware of the need to check information that's provided by a generative AI tool, something Slack can highlight in its app.

"There's so much that we can do with how we build the product to nudge people to check the results [of generative AI]," she said. Providing easily accessible edit functions would also help here, she said. In addition, LLMs will likely become more accurate as the technology develops, which could alleviate some concerns.

Ra