The core problem here feels less like “notifications” and more like attention routing. When multiple agents compete for attention, the real cost is often the context switch rather than the missed approval itself. I’m curious whether CrewTower could eventually learn which interruptions are actually urgent and surface only the ones that require immediate action. That could make the notch not just a notification layer, but an intelligent attention filter for parallel AI workflows.
Yes, exactly. I think “attention routing” is actually a better way to describe it.
With CrewTower, a notification isn’t just an informational update from an agent. It means the agent has reached a point where it genuinely needs your input and the session is waiting on you.
That’s really why I built it. I wanted to know exactly when an agent needs me, instead of constantly checking multiple terminal sessions.
CrewTower does this by listening to the agents’ hooks and detecting those moments where human action is required. So ideally, when CrewTower interrupts you, there’s actually something for you to do.
That makes a lot of sense. The interesting part for me is that CrewTower is not really adding another notification layer it’s reducing the uncertainty of when human attention is actually required. That distinction could become a powerful part of the product’s positioning. Curious how you’re thinking about measuring whether CrewTower is actually reducing context switches and agent idle time for users.
The “notice when the agent needs you” problem feels more fundamental than notifications themselves.
Curious whether the biggest productivity loss is actually missed approvals, or the context switching required to handle them.
For me, it’s definitely the context switching. I usually run three agents in parallel, and constantly jumping between terminals just to check whether an agent needs me is the bigger distraction.
That’s actually why I built CrewTower around the notch. I can keep all my agents at eye level, see their status, and only switch context when one of them actually needs my input.
That’s useful context. I’d be interested in continuing the conversation beyond the thread — would you be open to sharing the best email to reach you on?
About
I run three or four coding agents in different tabs and only notice a waiting one twenty minutes after it asked. Answering it costs me the work I was in. That is why CrewTower exists.
6 Comments
Yes, exactly. I think “attention routing” is actually a better way to describe it.
With CrewTower, a notification isn’t just an informational update from an agent. It means the agent has reached a point where it genuinely needs your input and the session is waiting on you.
That’s really why I built it. I wanted to know exactly when an agent needs me, instead of constantly checking multiple terminal sessions.
CrewTower does this by listening to the agents’ hooks and detecting those moments where human action is required. So ideally, when CrewTower interrupts you, there’s actually something for you to do.
For me, it’s definitely the context switching. I usually run three agents in parallel, and constantly jumping between terminals just to check whether an agent needs me is the bigger distraction.
That’s actually why I built CrewTower around the notch. I can keep all my agents at eye level, see their status, and only switch context when one of them actually needs my input.