The problem
Building with coding agents makes you the glue: copying handovers between Claude Code and Codex, reconstructing plans, picking a machine, reconnecting the same tools for each agent.
The approach
One workspace per project that holds the plan, the decisions and the memory. You direct the work in one conversation, on the web or in Telegram; it goes to the right agent on the right machine, and the results and the decisions that need you come back.
What we learned
The hard part is not running agents but continuity. The plan, the decisions and the exact code checkpoint have to travel with the work, or every switch starts from zero.
Next
Paired machines with authenticated runners, then safe dispatch and results, Claude working alongside Codex, and Telegram. Then an early-access alpha.








