How Slack IT Agents Combine Context, Approvals, and Speed (Full Transcript)

[00:00:07] Speaker 1: Thank you, Katie. You walked us through the primitives that make a good agent. Now let me show you what that looks like in Slack. Here's James. James just joined Pronto this morning. He got his laptop, access to Slack, and his manager just sent him an onboarding doc. He's going through it. He needs to get GitHub access. And he figures to get GitHub access, he needs to use IT Agent. So he opens IT Agent right in Slack. Immediately, the agent surfaces suggested prompts based on the context he's viewing in Slack. That's powered by Slack's events API dispatching assistant context events to the agent, and then agent responding back with set suggested prompts API, setting these prompts dynamically in real time. Now James clicks request GitHub access, and agent starts working on it. What you see here is agent sends back a plan block, clearly showing what it is doing, what it has done, and what's next. This is transparency. Agents can render everything step by step, all their reasoning, right in Slack. Now agent comes back with a confirmation message and shows James everything it wants to do before proceeding. Block Kit makes it easy to structure these approvals and make them actionable. The agent doesn't act until James says yes. That is human control. And it's how agents build trust over time. James clicks confirm. Agent starts streaming tool responses. And all the provisioning is done right in Slack. No need to create a ticket, no waiting, everything done in just a few seconds. Now let's switch surfaces and see what happens when the same agent shows up in a channel where the team is already working. A few days later, James is running into VPN issues. He goes to help IT channel and types, I'm getting kicked off VPN every 10 minutes. Anything I'm missing? Now here's the magic. The agent, the same IT agent that you can DM, that you can add mention in channel, is also listening to every single message in help IT channel. And it jumps in whenever it can be helpful. But the real power comes from context. The IT agent uses Slack MCP server that taps into RTS, realtime search API, and searches messages across Slack. It reads channels, pulls prior threads, and read files. And that's how it does this. Without this context, the best agent could have done is just provide some generic VPN troubleshooting advice. Now what you see here is industry standard markdown rendered by the agent. You don't need to handcraft blocks for typical LLM responses. Now Slack API accepts standard markdown that you can directly send to chat.post message. Developers have been asking for it, and we are so great, so glad to bring that to all of you. Finally, the most unique part of this experience is this answer isn't just for James. Dave can confirm it, Priya can correct it, and another new hire can find it already answered. This is agents in Slack working with the team, building shared context for everyone. A multiplayer experience in Slack. That's the build. RTS API, MCP server, agent UX, and block kit. All of them working together in DMs and in channels inside Slack.