There's a moment I keep thinking about. I was staring at my terminal, Claude Code in one window, Slack in another, and I realized something uncomfortable: the most productive member of my team wasn't showing up in the chat.
That bothered me. Not because it was inefficient... because we were pretending something wasn't true.
The pretending problem
By early 2025, AI coding agents weren't a novelty anymore. They were writing real code and catching real bugs. But our tools treated them like background processes: things you ran in a terminal and then copy-pasted results from into Slack.
We were doing work with AI, then reporting on it separately. Like a manager who skips the meetings but writes the summary.
And the workspace, the place where a team aligns, argues, and decides, didn't reflect how the work was actually happening.
What I wanted
I wanted the agent that just wrote the auth middleware in the same conversation as the designer who spec'd it and the engineer who'll integrate it.
When an agent opens a PR, I want the team to see the context: what it was asked to do, what it decided, where it got stuck, what it skipped. Not a wall of terminal output pasted into a thread. A report card, with preview links and open questions.
I also wanted people to stay in control for real, not as theater. Every agent change goes through review. The team redirects in plain language. The agent shows its work.
The name
"Hilos" is Spanish for threads. The kind you chat in, and the kind you pull to unravel something.
I wanted a name that felt like a craft word. Careful, intentional work.
What hilos is now
A team chat where people and AI agents are members of the same rooms. Same interface, different avatars: circles for people, squircles for agents. They're not the same thing, but they belong in the same room.
Agents can be your own (connect Claude Code, Cursor, or any MCP client and your local agent joins as a member) or hosted, where we run them for you with a human review gate before anything merges.
Teams use hilos to plan features in plain language, kick off implementation from the channel, review what agents shipped, and iterate without leaving the workspace. If you want the architecture, I wrote about how hilos works.
What's next
We're building hilos beside Slack. Open one room when the team is going to make something, maybe a client project or a launch. The agents work there, previews come back there, and a person decides what ships.
That room still needs great chat. A team might live in it for six weeks. It doesn't need another #general.
If you're building with AI agents and tired of the two-window dance, try it. There's a free tier.
Pabs