Main content

Slack matures as a workflow automation platform, juiced with native AI

Phil Wainewright Profile picture for user pwainewright September 6, 2023
Ahead of Dreamforce next week, Slack launches Slack AI, based on its own native LLM, and adds finishing touches to workflow automation, including a Lists feature for structured tasks, approvals and information.

Slack Workflow Builder Automation Hub
Slack Workflow Builder Automation Hub (Salesforce/Slack)

Slack is stealing a march on next week's expected flurry of Dreamforce announcements from its parent company Salesforce, today unveiling Slack-native generative AI capabilities, a new lists function for structured workflow, and updates to its automation platform. While the AI news is bound to grab the limelight, the automation platform is the true star. Today's announcements, coming on the heels of a UI refresh announced last month, complete Slack's long-running metamorphosis from a messaging platform to a complete workflow automation proposition. Ali Rayl, SVP, Product Management at Slack, comments:

It's become so much more than a messaging platform. One of the main reasons behind the new interface is to make sure that all this power is much more close, at people's fingertips. We had a very messaging platform centric UI, because we developed that UI when we are a messaging platform, and we offer so much more now. So one big question for us is just, people have all of this, how do we make it more obvious and useful to them so that they get more value out of the product?

Slack AI goes native

The new Slack AI brings its own dedicated Large Language Model (LLM) natively into Slack, enabling features such as summarizing a message thread or recent activity in a channel, while adding natural language questions and answers in search. These capabilities go into limited pilot with customers later in the year. This is an evolution since the Slack GPT announcements at World Tour New York, as Rayl explains:

Since that time, we've really been digging in on a lot of different possibilities with LLMs and asking ourselves, what can we do that brings the most value to our product and to our customers quickly? We've run through nine or 10 different experiments with LLMs and this is where we landed, the first opportunities that we have to bring AI into the product ...

This state of AI functionality, we're powering it with an LLM that we're running ourselves. What this means is, it's not calling out to the Internet, it's not calling out to some company's LLM that's running in their own data centre, it's not being used to train anybody else's LLM. This is just native in Slack, what can we run to make our own product better?

Rather than being a separate menu item in the new UI, Slack AI will bring a set of capabilities that are embedded throughout the product where they're most relevant. During the pilot, product managers will be fine-tuning these capabilities to make them as effective as possible for users. She goes on:

A lot of what we're doing right now is prompt engineering. Figuring out exactly what apps should we use for exactly the kind of data that Slack has, to give the best answers possible, the best summaries possible.

Another important facet of Slack AI is attribution. Unlike consumer-facing generative AI products that give an answer with no indication of what sources were used to generate it — answers that may be based on false information or even made up or 'hallucinated' — Slack AI will quote its sources so that business users can verify the information or dig deeper. She explains:

One thing that we've decided is really important for confidence and accuracy is attribution. You mentioned hallucinations. One thing that you'll see in the summaries [is] attribution — who said it — and number two, a link to more. Sometimes the summary isn't enough, I need to dig into this conversation. But we want to make sure that everything that we show you, you can have confidence in because you can tie it to where it came from and who it came from.

Structured task management with Lists

Moving to the automation capabilities in today's announcements, a new feature called Lists is hugely significant. Again due to pilot this winter, with general availability following next year, this provides a structured format for tracking tasks, projects and approvals. Previously, users have had to use third-party apps integrated with Slack for this kind of structured tasks management. With Lists becoming a native capability within Slack — available as one of the tools in the new 'create' menu introduced in the updated UI — users can use all of its messaging capabilities to collaborate more easily around these tasks and approval flows. Rayl elaborates:

Today, Slack has a lot of information. But it's people writing messages and channels. The good thing is, information is organized in channels. Information is highly readable, because it's people typing sentences. But one of the drawbacks is that it's not structured. And a lot of the work that we do in every industry, every kind of job, is more with structured data. This is why spreadsheets are so popular ...

As we've looked at Slack, and as we've looked at how people work, and what else can we do — what else can we provide to make work better for folks? — this has really stood out as a gap that we wanted to solve, which is how can teams not only bring structured data into Slack, but also collaborate around it and have conversations around it? Because ultimately, the heart of Slack is the conversation, it is working together, collaborating, making meaning of data to move work forward.

Lists can live inside of Slack message threads, channels, or a canvas, the collaborative workspace introduced last year to provide a shared placeholder for links to relevant documents, data and workflows. Lists can also be included in workflows, for example when the output of a workflow is written to a list, and lists can contain a canvas. Given the overlap in some of their capabilities, I asked Rayl to clarify the difference between a canvas and a list. She explains:

Canvas, think of this more as a flexible surface for whatever — long-form text, who's on the team, workflows. It's a space to keep track of everything that isn't a channel where everything scrolls up. It's more of a resting home. Lists, on the other hand, [are for when] teams need to structure data as records. Within a list, every single one of these rows is a record ...

Exactly like everything else in Slack, you can at-mention people, it'll generate a notification, it'll show up in your threads, you can share these messages to different channels, you can pull them out, whatever you need. What we're trying to do here is to bring a real collaboration-first mindset to the more structured data needs of teams. That collaboration-first mindset also exists in canvas, but the data doesn't have to be as structured.

Workflow automation for all

With all of these extra formats where information can live within Slack, search is growing in importance as a means of surfacing content from wherever it's located, so beefing it up with the addition of Slack AI is timely. She adds:

That's not only searching in messages, it's also searching your canvases, it's also searching your lists. So we're going to fan out to do our best to surface the right answer for you, no matter where it happens to live in Slack, because Slack is becoming bigger, and things live in more places. And so it's incumbent upon us to help navigate, simplify, make it easier to discover all that data.

Some finishing touches to workflow automation and the Slack platform round out today's announcements, along with a new hub for discovering automations, and editable templates to help get people started. Automations also becomes a new addition to the same 'create' menu that Lists will join when it becomes available. Rayl says Slack has been following three tenets while developing its platform capabilities. First of all, the workflow builder should be available to all. She comments:

A lot of the folks responsible for business processes can't write code, and that's okay. This is the gap that we're trying to close with all the new platform functionality we've been releasing over the last year.

Secondly, it should connect across Salesforce, via Salesforce Flow, and have connectors out to other products, whether that's Google Sheets, JIRA or whatever.

Finally, there's the ability to host custom code, which used to require spinning up a separate server but a few months ago became possible to host natively on Slack and make it accessible in the workflow builder. She explains:

Imagine a scenario where you need a custom workflow to pull data out of a specific system. You can have an engineer write that once. You can make that function available to everybody in the company. And even people who don't know how to write code can then pull in that function, and use that to build their own workflows.

This is the ecosystem we're working on. How do we make lives easier for developers? How do we make it easier for system administrators? And how do we make it accessible to more people in any role in the company?

My take

When Slack announced its UI update last month, I speculated that it might be holding back some AI news for Dreamforce week. Well, the news has arrived, but one week in advance. On reflection, that makes sense for Slack as Dreamforce is really very much about the Salesforce ecosystem and customer base, whereas Slack plays in a slightly different competitive landscape. Making a separate announcement that doesn't get lost in all the Dreamforce noise gives it more visibility in the competitive context of what what is coming out from rival messaging platform Microsoft Teams and all of the many other workflow automation and digital teamwork.

On that note, Slack really has matured as a full teamwork platform. When it first announced its platform beta last year, I said Slack had come of age as an enterprise digital teamwork platform. The addition of Lists now takes it to another level, filling a key gap in its no-code workflow automation capabilities. What started out as solely a messaging platform is now very much about organizing work and automating processes with the underlying messaging infrastructure becoming part of the platform for hosting all of this other activity that forms part of people's day to day work. It's one more step along the path to fulfilling the vision that diginomica calls the Collaborative Canvas — a shared fabric for channeling enterprise teamwork.

A grey colored placeholder image