How We Rebuilt Slackbot

Slackbot has been part of our DNA since the very beginning. For years, it served as your friendly notifier, offering a helpful, if somewhat limited, guide to the basics of the platform. But as the world of work has evolved, so has our vision for what a bot can be.

This month, we introduced the all-new Slackbot: your personal AI agent in Slack. It’s a transition from a basic notification tool to a sophisticated agent that lives with you in the flow of work. There is nothing to install, learn, or manage; it just works. 

Across Salesforce, more than 42,000 employees are already using Slackbot to save a combined 138,000 hours per week, translating to $6.4 million in productivity gains. Top sales adopters are saving up to 20 hours per week individually. With 96% user satisfaction, the highest rating in Salesforce history, this proves that context-aware AI works at scale.

Our goal, beyond making Slack more powerful, was to bring the same kind of simplicity, courtesy, and magic that made people fall in love with Slack in the first place. By combining modern AI with our unique brand personality, we’ve designed a tool that helps you work more intelligently than ever before and makes work fun again.

Building inside a community

The journey to the new Slackbot didn’t start with a top-down mandate. The spark was a weekend side project, which quickly gained momentum that became impossible to ignore.

We guided this development with one of our core product principles we call “prototyping the path.” Instead of spending months planning the perfect tool on paper, we started building and using it immediately within our own flow of work, even when the experience was still a work in progress. We weren’t thinking about perfection but, rather, about how to build a tool people would want to use again. That mindset gave us permission to ship early, listen closely, and iterate based on real behavior, not assumptions.

Building Slackbot like this was a powerful lesson in how much you can learn by putting an early idea in front of real users and watching what actually happens. Internal teams and pilot customers taught us where Slackbot clicked, where it didn’t, how people discovered it, and what made them come back. We were looking for glimmers: those specific moments where the AI felt genuinely helpful and intuitive. Then we chased them.

This was a team effort. We launched a channel for Slackbot feature requests that quickly grew to thousands of members across Salesforce. This community became our ultimate testing ground, providing real-time feedback that allowed our engineers to iterate and implement changes in days rather than months. The feedback we received was key to our early iterations, but when Slackbot was internally released to all of Salesforce, we benefited from hearing the opinions of the tens of thousands of coworkers who were using it every day. From there, we extended the pilot to a group of existing Slack AI customers and kept iterating based on their input. We built this tool for our users, but more importantly, we built it with them.

Context: the technical secret sauce

To power this experience, we chose Anthropic’s Claude, a frontier model that provides the high-level reasoning required for a true personal assistant. But the real magic of Slackbot is in its context engineering.

For an AI to be useful in a workspace, it needs the right data at the right time. When you consider the fact that your business has collected historical data in Slack, there’s no better place for an employee agent to live. But the key is that context continues to evolve: Slackbot works to understand you as a user. It considers who you’re working with, the projects you’re working on, who you’re interacting with, and where your attention is going in Slack. That’s what makes it feel so personal.

With this context in mind, we integrated Slackbot deep within our existing search infrastructure. Our search tech is incredibly robust, and Slackbot uses it to find and synthesize information across your entire workspace faster than any human could.

Because Slackbot is a native component of Slack, it has a “surface awareness” that gives it a deep understanding of the context of your conversations, canvases, and lists, and the specific way you interact with your colleagues. This native integration allows it to act as a seamless layer over your work, instead of another app you have to manage.

“Thinking” like a human

When we first began this project, Slackbot mostly worked with basic natural language searches. It was a good start, but it wasn’t enough for the complexities of modern work. We knew we needed to make it more precise by engineering a few things:

• Advanced filtering: We enabled Slackbot to use the same sophisticated filters you use in Slack’s search bar, such as specific date ranges and “from” parameters. This ensures that it pulls only the most relevant information into its context window.
• Parallel and iterative searching: This was our biggest technical unlock. Instead of doing one search and stopping, Slackbot can now attempt multiple related queries simultaneously. If it doesn’t find the answer immediately, it builds on previous results to dig deeper, just like a human researcher would.
• Guided intelligence: We designed Slackbot to be more than just a good host. We want it to serve as a warm and infinitely knowledgeable coworker that’s always there for you. If a user provides a weak prompt or lacks AI fluency, Slackbot doesn’t hit a dead end. It will propose alternative solutions or suggest follow-up questions to guide the user to the information they need.

One key part of Slackbot is that it “shows its work” and links to its sources. When Slackbot shares its reasoning (e.g., explaining why it’s searching certain channels, acknowledging what it doesn’t know, or walking through how it arrived at an answer), it transforms from a black box into a thinking partner. 

This transparency builds confidence in Slackbot’s outputs, and it models the kind of clear communication we want to see in every workplace interaction. By “thinking out loud,” Slackbot teaches users not just what it knows but how it knows it, and, in doing so, invites them to think more critically alongside it.

Making Slackbot intelligent, helpful, and fun

Efficiency is only half the battle. We wanted Slackbot to feel distinctly “Slacky.” We iterated extensively on its system prompt to make its voice congruent with our brand: intelligent and helpful but also completely approachable and even a little bit fun.

We’ve baked in delightful details that create a genuine connection between the user and the tool:

• Affinity awareness: Our search ranking considers your rapport with other people and channels. It knows which conversations matter most to you, making its results feel deeply personalized.
• Tone matching: Slackbot can implicitly match your vibe. If you’re being all-business, it stays professional. If you joke with it, it’s designed to joke back.
• Custom touches: From using your own most frequently used emoji to the occasional heart-eyes in the top bar, we wanted to bring back some of the whimsy of early Slack.

Here’s an example: When you ask Slackbot what kind of animal you’d be, and it just gets you based on your recent work, it stops feeling like a transactional chatbot and starts feeling like a teammate. That emotional connection is a unique value proposition that makes you feel smarter and more supported at work.

Trust and security

Slackbot is built on the same foundation of trust as Slack itself. It’s protected by Slack AI Guardrails, a multi-layered safety framework designed to keep your data secure. It respects roles, permissions, and access controls, and only surfaces information you are authorized to see, with clear citations that show where information comes from. 

Your interactions remain private to you. Your data is protected and handled in accordance with Slack’s security and compliance standards. Real-time safeguards detect and defuse issues like prompt injection, unsafe content, or phishing attempts, while strict permission checks ensure that Slackbot only accesses data that users are already authorized to view within Slack. 

This is enterprise-grade trust and security that customers expect, delivered through a personal, conversational experience.

The new mental model

When I think about how Slackbot will change the way we work, I don’t think about a single feature or capability. I think about a shift in how people approach their workday. We know that employees spend a lot of their time on the “work of work,” where they’re constantly looking for files, trying to remember what was said in a meeting, or catching up after time off. Slackbot cuts right through all of that and multiplies your productivity.

It becomes the place people go to understand what’s happening, decide what to do next, and move work forward. That’s the real change: not simply doing work for people, but reducing the friction between understanding, decision-making, and action at scale, without making work feel more complex.

Looking ahead, we see Slackbot playing an even larger role in taking action on behalf of users as it becomes even faster, more capable, and, above all, in sync with employees and other agents in the Slack platform. As our ecosystem of AI tools expands, Slackbot will become the front door to the agentic enterprise, the first place you go to start working across Agentforce, Anthropic, Linear, OpenAI, Writer, and other third-party agents.

We’ve designed Slackbot to make the workplace feel intuitive, efficient, and, most importantly, delightful again. We can’t wait for you to start teaming up with your new favorite work friend.