Slack is turning into an AI agent hub. Should it?

The head of Slack, Denise Dresser, tells TechCrunch she is shifting the business chat platform into a “work operating system,” specifically by making Slack a hub for AI applications from Salesforce, Adobe, and Anthropic. The company’s CEO sees Slack as more than a place to chat with your coworkers, but do users want that? And if they do, will they pay a premium for it?

Slack announced several new features on Monday for a pricier tier of the messaging platform: Slack AI. The updates include AI-generated Huddle summaries, similar to the channel summaries already available to those subscribers. Users can also now chat with Salesforce’s AI agents in Slack, alongside tools from third parties that will enable AI web search and AI image generation.

Salesforce purchased Slack in 2021, shortly after the messaging platform became a staple of remote work for millions of people. Three years later, Salesforce is pivoting hard to AI agents — apparently so hard that their popular messaging service is doing it too. Slack CEO Denise Dresser says the platform will play a key role in the transformation, since it’s a natural place to interact with AI agents, because people are already chatting there throughout the workday.

“AI is showing us a new way to experience technology which is very organic to Slack: it’s conversational, you’re surfacing information, and you’re taking action right in the flow of work,” said Dresser, who took over as Slack’s chief executive 10 months ago, in an interview. “There’s probably not a better place and product than Slack to allow you to do that.”

But why does Slack need AI? Ever since ChatGPT launched in 2022, a lot of companies have introduced AI features as a way to appear “cutting edge” even if the integration doesn’t make much sense to the core product. Slack adding AI agents to its messaging service doesn’t seem to be an obvious exception.

Dresser’s justification for AI agents is that Slack is not simply a work messaging platform, but rather a digital workplace or work OS that “brings all your people and processes together.”

The head of Slack tells TechCrunch that every CEO is asking for AI features, such as ways to quickly catch up on team discussions or tools to surface information buried in some database. These are some small ways Slack is trying to bring companies into the AI era, she explained.

One of Slack’s new agents, Agentforce, will allow Salesforce customers to perform on-demand analysis of business data directly in Slack. Cohere and Anthropic’s Slack agents will offer similar services, as long as you’re paying for their enterprise AI services.

Perplexity is also releasing an agent for Slack that will let you search the web. Adobe Express’s Slack agent will let you create branded content from text prompts within the messaging service.

Klarna’s CEO made headlines last month when he announced plans to drop Salesforce and Workday as software providers and replace them with internally built AI tools. Andreessen Horowitz partners published a blog post in July predicting changes like this, where businesses would shift away from the pricey CRM services in favor of in-house AI solutions. Salesforce CEO Marc Benioff is skeptical of Klarna’s AI solutions and wants to see some proof the company is really doing it.

When asked about the Klarna CEO’s comments, Dresser said that enterprise AI solutions need to be trustworthy and safe, two things Salesforce is trying to guarantee for customers.

That trust was tested earlier this year when Slack came under fire for training a recommender system on customer data by default , according to part of its privacy policy found by developers on Hacker News. It was later revealed that Slack was using customer data to power emoji recommendations and was not using it to train its large language models underlying Slack AI. But still, the privacy policy claimed that Slack required users to send an email to the company if they didn’t want their messages to be part of Slack’s training data.

Slack claimed to not use customer data to train Slack AI then and continues to today.

“No LLMs are trained on Slack data, period,” said chief product officer Rob Seaman in an interview with TechCrunch. “Honestly, there was a hiccup and an update to a policy on our website that we could have handled better. Especially in this age of AI and heightened awareness of how your data is being used, that turned into a thing we wish hadn’t.”

These questions around privacy only become more prevalent as Slack leans further into AI. The service is changing from strictly messaging into one where AI tools are pulling information in and out of the platform. Users have good reason to be skeptical as AI becomes just another tool in the box.