I’ve been using Codex and Claude a lot this month. I’m on ChatGPT Pro ($200/month) and Claude Pro($20/month).
One of my main workflows is to have Claude plan the work, assign tasks to Astra, and keep things moving toward a longer-term goal. I’d really like Astra to handle that itself: plan, delegate to subagents, run background terminal jobs, and follow through.
I actually prefer the ChatGPT desktop app overall, which makes this frustrating. In my experience, Astra often spends too long digging into the what and why of every step, including repeatedly checking background jobs when nothing has changed. That takes time and fills the context with low-value updates. On longer tasks, too much effort seems to go into checking and rethinking instead of carrying the plan through. I still end up relying on Claude to keep the overall workflow moving.
Two changes would really help:
-
Make background jobs visible, like subagent tasks: show the command, status, elapsed time, and latest or final output in a dedicated panel.
-
Let the process run, then bring the agent back. When a job finishes, fails, or meets a defined output condition, automatically resume the agent in the same conversation with the result. The app should handle routine waiting and checks under clear rules, without repeatedly waking the model just to check whether anything changed.
I want to hand over a goal and see it move forward, with enough visibility to step in when needed.
Attached: Claude’s background task UI running a Codex command.
