Most of my time building feels like a dance across (i) Claude/Codex TUIs (ii) Github PRs and (iii) Jira/Linear/etc issues. The greater the parallelization, the more intricate the dance feels. At some point, parallel work on N issues leads to 3*N windows. OS window management starts to feel cumbersome and ultimately constraining. You can go a bit further with things like using different desktops, one per project, and at least only seeing the windows of one project at a time.
I’ve looked into agent orchestration tools like Herdr, Emdash, Conductor, Superset but felt features most useful to me were spread across them. Herdr is built more around agents and panes and less around issues. The rest may come closer with linking PRs and issues, but they’re desktop apps. Claude Code’s agent view is good for hopping between sessions, but it doesn’t currently address the window overload problem.
My preference for the Claude/Codex TUIs over the desktop versions nudged me to explore what blend of the two I can get away with. This led to a Rust shell where the issue and PR sit beside a Claude/Codex session. It’s a TUI with its prefix key etc. that also supports less traditionally-TUI things like drag-to-reorder for projects/issues. It’s a Claude/Codex++ designed around multiple sessions and minimizing chasing windows or alt-tabbing. Ultimately, it remains an extension over my Claude/Codex TUI experience and helps with the multitasking dance across different live threads. Curious if this approach or even problem statement resonates with others.
Happy to answer questions and feedback is very welcome!
On the intuitive bit... TUIs do feel more immersive (if that’s the right word), enough to make the software feel more like an extension of my thought process than something outside it.
Perhaps taskman could work inside Warp, haven’t tried this yet :) Tabs were my setup before, but it didn’t scale well for me: for every terminal tab I ended up with ~2 browser tabs e.g Github PR and Jira/Linear issue with comments.
Curious about the number of threads/tasks you juggle at the same time until the context switch costs take over.
Most of my time building feels like a dance across (i) Claude/Codex TUIs (ii) Github PRs and (iii) Jira/Linear/etc issues. The greater the parallelization, the more intricate the dance feels. At some point, parallel work on N issues leads to 3*N windows. OS window management starts to feel cumbersome and ultimately constraining. You can go a bit further with things like using different desktops, one per project, and at least only seeing the windows of one project at a time.
I’ve looked into agent orchestration tools like Herdr, Emdash, Conductor, Superset but felt features most useful to me were spread across them. Herdr is built more around agents and panes and less around issues. The rest may come closer with linking PRs and issues, but they’re desktop apps. Claude Code’s agent view is good for hopping between sessions, but it doesn’t currently address the window overload problem.
My preference for the Claude/Codex TUIs over the desktop versions nudged me to explore what blend of the two I can get away with. This led to a Rust shell where the issue and PR sit beside a Claude/Codex session. It’s a TUI with its prefix key etc. that also supports less traditionally-TUI things like drag-to-reorder for projects/issues. It’s a Claude/Codex++ designed around multiple sessions and minimizing chasing windows or alt-tabbing. Ultimately, it remains an extension over my Claude/Codex TUI experience and helps with the multitasking dance across different live threads. Curious if this approach or even problem statement resonates with others.
Happy to answer questions and feedback is very welcome!
I also like TUIs—they just feel more intuitive to use. I usually open multiple tabs in Warp.
On the intuitive bit... TUIs do feel more immersive (if that’s the right word), enough to make the software feel more like an extension of my thought process than something outside it.
Perhaps taskman could work inside Warp, haven’t tried this yet :) Tabs were my setup before, but it didn’t scale well for me: for every terminal tab I ended up with ~2 browser tabs e.g Github PR and Jira/Linear issue with comments.
Curious about the number of threads/tasks you juggle at the same time until the context switch costs take over.