A few months ago, I started experimenting with building AI agents as Slack bots. What began as a fun weekend project has now evolved into a core part of how our team works every day. These bots live in Slack and handle real, non-trivial workflows such as:
- When a new signup comes in, one of them enriches the lead via Apollo, drafts a personalized email, and drops it into my mailbox.
- If someone mentions a new task in any channel, the bot jumps in, asks a few clarifying questions, and files a Linear ticket.
- Collecting customer info from Stripe for transaction failures and bringing it into the slack thread automatically.
- Scouring the web and drafting blog posts - sure you can do this with ChatGPT. But doing it collaboratively on Slack gives a different overall experience.
- We even made them proactive. For example, if someone talks about sprint planning, the bot jumps into the thread and asks if it should help break things down into tasks.
At first, this felt like "just automation." But what surprised us was how collaborative it became.
Since the agents live in public Slack threads, teammates started observing how others interacted with them — how people phrased prompts, tagged bots, corrected outputs. It turned into a sort of lightweight, social learning system. Everyone got better at prompting and thinking in “agentic workflows” just by being part of the conversation.
We also built a lightweight dashboard to help anyone on the team spin up and deploy new bots, manage access, customize behavior, and attach tools.
Demo: Here’s a short video of one of our bots in action — it self-corrects, navigates tools like Linear and internal APIs, and finishes the task end-to-end.
Stack:
- Vercel AI SDK
- axllm.dev(DSPy for typescript) for the agentic loop
- Composio for tool integrations
We are now running a closed beta for anyone interested in deploying Slack bots as agents. If interested, join our waitlist - https://heyzest.ai/