The office chat that replaced email is now also where AI agents read what's actually happening inside a company.
In practice, Slack replaces both email and the conversation you'd have leaning over a colleague's shoulder. Instead of writing 'Hi, hope you're well', you open a channel named after the project and drop a single sentence. Everything stays in one place, searchable, with history going back. The team at CODE, my agency, worked exactly like that for years, before we moved to our own tools.
The product started in 2013 as a side effect of a company that made a computer game and went bankrupt. Founder Stewart Butterfield - a Canadian entrepreneur who had earlier created Flickr - noticed that the internal chat his team had built for itself was worth more than the game. He turned it into a standalone product, and it took off like a rocket.
The owner today is Salesforce - a large American software company that helps businesses track customers and sales (a system called CRM). It bought Slack in 2021 for $27.7 billion, because it needed a way into companies' daily communication, not just their databases.
Here's the part that makes this news at all. Slack now has built-in AI that reads a channel's history and summarizes what you missed, and companies connect agents to it - automatic programs that follow conversations and carry out tasks on them without a person pressing a button. The channel where you used to chat with a colleague now has a third participant. Nobody invited it personally - it's on by default.
For me, this is where the quality shows
Slack started as something modest - a chat so replies between colleagues wouldn't get lost - and turned into something a big corporation paid nearly $28 billion for, because it knows where a company's real life happens. Not in the presentations. In the channel where someone writes 'this client still hasn't paid'.
That's why I now look at the second layer a bit more carefully. When a channel for client complaints or an internal argument also has an AI reading and summarizing it, the line between 'internal conversation' and 'data training someone's model' is much thinner than it looks the moment you type the message. Remember that the next time you hit Enter.