Conduit began as a shared workspace for AI agents. I wanted Cursor, Claude, and Codex to collaborate inside one platform, with a common place for their work.
One idea mattered more than the interface: the Seat. An agent or session could change, but the Seat would retain its role and work history. A replacement agent should be able to continue from the work already done, rather than reconstructing it from a conversation.
The idea was clear. The scope was too large. I kept spending my days fixing bugs, without getting much closer to a tool I could actually use.
Find the part that already works
I split Conduit into smaller pieces. The first became Work Stack: a place to organize goals, tasks, and the records needed to continue them.
That part was familiar. I had already been using agent skills for backlog management, worklogs, weekly reports, and OKRs. Connecting those workflows gave Work Stack a concrete starting point. I was building around a working habit, with a problem I could recognize in daily use.
Before asking agents to implement it, I described the philosophy and asked for HTML mockups as if the finished product already existed. I refined the examples, chose one, and used it as the starting point for implementation. Improvements followed as I used the app.
Kanban tools and Super Productivity were useful references. They helped me explore existing approaches, while my own workflow kept guiding what to bring together and what to change.
Recording work leaves another problem
Work Stack helped me organize my work and miss fewer important tasks or schedules. I now use it in both work and everyday life. But my personal projects exposed another gap: registering a task did not make me return to it.
I was already using multi-agent tools I liked. What I wanted next was a closer connection between a recorded task and the agents that could work on it. That became Stackbite, the execution layer growing alongside Work Stack.
The current connection uses a CLI/SDK bridge. The full flow I want—selecting a task, forming a team, and starting work from one place—is still a goal. The project page describes what exists today and what remains ahead.
Keep the larger idea, make the next piece useful
Conduit remains the long-term direction: persistent work context connected to execution. Work Stack and Stackbite let me explore that direction in pieces I can use and improve.
Breaking the project apart gave me a more practical next step. I could see whether a smaller part helped me, discover its limits, and let those limits shape the next tool.