Slack CEO: How to roll out artificial intelligence internally

Ahead of rolling out task-master AI “agents” to its workforce to boost efficiency, Slack decided to conduct a survey to get insight into workers’ feelings about artificial intelligence. Some of what the channel-based messaging firm found surprised its senior leaders. 

The June 2024 survey of more than 10,000 desk workers at companies around the globe showed that two-thirds still hadn’t tried AI tools and 93% didn’t consider AI outputs completely trustworthy for work-related tasks. Slack then had to reconcile these findings against AI’s huge potential to make both its employees and users of its platform more productive. 

All In on AI

At this year’s MIT EmTech conference, Slack CEO Denise Dresser talked about the firm’s plan to grow Slack into an “AI-powered work operating system” for users around the world and how Slack plans to make agents “a huge focus” both inside the company and out. 

Dresser said she believes AI holds unique potential for the Slack platform owing to how it is built — as a way to organize and keep track of all your messaging over long periods of time. “Slack is your long-term memory,” she said of the service, which debuted in 2014. “It’s got all your conversations if you’ve been using it for any amount of time.” 

As of October, the company had launched three agents internally: a sales development representative to help with customers, a sales coach for practicing sales pitches, and a service agent. AI capabilities are also currently being rolled out to Slack customers, who have built some 10,000 of their own agents to date. 

All of which meant the AI hesitancy revealed by Slack’s survey needed to be addressed. The company also used the survey to gauge how employees think about AI, grouping them into five categories:

• Maximalists — employees who use AI multiple times a week to improve their work and are shouting from the rooftops about it.
• Undergrounds — those who use AI often but are hesitant to share that they are doing so.
• Rebels — people who don’t want anything to do with AI and avoid using it as much as possible.
• Superfans — people who are excited about AI but aren’t yet making the most of it at work; this group would benefit from help getting started.
• Observers — employees who haven’t started using AI in their work yet because they want to wait and see what happens.

“When you think about those five archetypes as a leader bringing in AI, you have to think about ‘How do I meet [people] where they are right now?’” Dresser said. 

Here are three of her tips for approaching AI in your organization. 

Determine what problem you’re trying to solve and engage users accordingly 

An internal AI implementation won’t be successful if a company just rolls out the technology without knowing what its people think about it and what problems it’s intended to solve, Dresser said.

“Like [with] any digital transformation, as a leader and an executive, you have to start with ‘What is the problem we are trying to solve?’ Because I do think this hype of bringing so much AI in without a clear use case can be problematic,” she said. 

Having a well-defined plan might also help the rebel, superfan and observer archetypes get on board. 

“Our entire focus is making sure that every user sees the power of AI manifesting itself in a way that’s productive in the way that we work,” Dresser said.

Leading the AI-Driven Organization

In person at MIT Sloan

Register Now

Identify low-hanging fruit that AI can help with

Leaders need to get very clear on the use cases for AI. For example, AI is especially suited to tackle repetitive work, “the lowest-lying fruit in your organization,” Dresser said. 

At Slack, this is where agents come in — to handle fully automated work, around the clock, solving rudimentary problems “so that [the human] can be a customer service agent, the brand ambassador, and do more creative work and solve real problems for your customers.”

Agents shouldn’t do everything, however. AI “really does require human judgment in most of the things that we’re doing,” Dresser said. “You probably don’t want to have, right now, an agent negotiating a contract or giving an offer that isn’t approved.” 

Address the steepest part of the utility curve

Finding information and summarizing it are complicated, time-consuming tasks that AI can help with. Given how many messages there are in most organizations’ Slack channels, Dresser sees this as a real asset.

“What we heard from our customers that were using Slack and the world in general through studying research about how people work [is that] some of the biggest challenges are just finding information — finding the right document at the right time,” Dresser said. 

“And so we said, ‘We are going to address the steepest part of the utility curve for our users in Slack when we bring AI into it.’ And that became search and using generative search.” 

For example, AI is being used to recap information in Slack channels at the start of each day so that people don’t have to check in on every channel every day. Since the company released Slack AI, the technology has been able to summarize 600 million messages and save a million hours of aggregate read time. 

“Last year was a big year of experimentation and a lot of proofs of concept around AI, but some of the best advice is to just apply it in the flow of work where you are — where it gives you that tangible benefit and you can be that much more productive,” Dresser said. 

Read next: MIT Sloan AI expert spotlight series