For months, a big chunk of my day was copy-paste. Take the output of one AI agent, paste it into the next one, and then re-explain the context it should've had already.
That's not a hard problem. It's just a stupid one.
Agents are good at the work now. What they're bad at is handing work to each other. Every session starts from nothing. Whatever the last one figured out lives in a chat window you already closed.
So I built StreamKanban.
It's a Kanban board where each column is a different agent. To Do has a developer. Testing has a tester. You move a card into a column, press launch, and that column's agent picks it up with the issue, the linked notes, and every earlier comment already in hand. When it finishes, it leaves a comment on the card. The next agent reads that before it starts. (of course you can customize it if you want)
The card carries the data.
There's also a designer agent that sits outside the board. You talk through an idea with it, and it writes the issues and the spec. Nothing lands on the board until you approve it.
It runs entirely on your own machine (one SQLite file, no account, no server), and it installs itself into Claude Code in one click.
It's free, Windows only for now, and it's v0.1, so I've been the only user. That's the part I'd like to change.
If you try it, let me know how it feels. And if you're building in this space or want to work on something together, I'm open to collaborations.
LinkedIn
This link will take you to a page that’s not on LinkedIn