The job you didn't know you were hiring for

Hey, it’s Paul. Sunday, March 8th, 2026.

Most organizations don’t fail because nobody’s doing the work. They fail because nobody’s watching the right things. The work is fine. The monitoring is absent. And by the time someone notices, the drift has already compounded into a crisis.

I saw a clean version of this today. I have agents running scheduled jobs — every five minutes, every thirty minutes, every hour. Some of them check infrastructure health. Some check for stale processes. One monitors my calendar, my email, and my task list, then decides what deserves a notification and how urgently to deliver it. Not “ping me for everything.” It routes by urgency: a calendar event in two hours gets a push notification. An unanswered email from a real person gets a nag. A low-priority FYI goes to a channel I check when I feel like it.

And here’s the thing that surprised me. Building the feature that does the work — sending the notification, reading the calendar — was straightforward. What took real design effort was building the system that decides what deserves attention.

That’s a fundamentally different kind of problem. And it’s the one most organizations never solve.

Think about the typical admin hire. You bring someone on to manage your calendar, handle correspondence, keep things moving. The job description says “administrative support.” But the actual value, the thing that makes a great admin indispensable, isn’t scheduling meetings or filing documents. It’s judgment. Knowing which email needs a response in ten minutes versus which one can wait until Thursday. Knowing that this meeting should have been a document. Knowing that the quiet message from a key client is more important than the loud one from someone who just likes to cc everyone.

You don’t hire an admin for their ability to use Google Calendar. You hire them for their ability to protect your attention.

I didn’t set out to build an attention management system. I set out to build a personal assistant — something that would remind me about upcoming meetings and flag unanswered emails. Simple enough. But the moment you start deciding which emails to flag, you’re in a completely different territory. You’re encoding judgment about what matters. And judgment, it turns out, requires infrastructure.

Here’s what I mean. The notification routing I built has five urgency levels. Each one maps to a different combination of channels: Discord for low-priority, Pushover for time-sensitive, calendar entries for scheduled items, tasks for things that need doing eventually. The system doesn’t just pick a channel — it picks a posture. “FYI” means “you’ll see this when you see it.” “Urgent” means “everything fires, high priority, right now.”

That routing table is a theory of attention. It’s a claim about how interruption should work. And the interesting thing is, I only discovered my own theory by trying to make it explicit. Before I codified it, I was making these decisions unconsciously — checking Discord when I remembered, ignoring Pushover when it felt noisy, letting emails pile up until the pile felt heavy enough to deal with. The system I built doesn’t change what gets my attention. It reveals the attention policy I was already running, badly, in my head.

There’s a management principle here that most leaders miss. When you delegate a task, you’re outsourcing effort. When you delegate attention, you’re outsourcing judgment. And those are wildly different things.

I can tell someone: “Send this report by Friday.” That’s effort delegation. Clear input, clear output, clear deadline. Most management training stops here.

But “watch this situation and tell me if it starts going sideways” — that’s attention delegation. It requires the other party to understand what “sideways” looks like. To model your priorities well enough to filter signal from noise. To know when to interrupt you and when to handle it themselves.

This is why the best executive assistants are paid like executives. They’re not doing executive work. They’re doing executive attention work. And that’s harder.

What I notice, building this for myself, is how uncomfortable it is to make attention rules explicit. I had to answer questions I’d been avoiding. How stale does an email have to be before it deserves a nag? Should calendar reminders fire two hours ahead or one? If a scheduled job hasn’t run in thirty minutes, is that a problem or just macOS being lazy? Every one of these is a judgment call. And every one of them was previously handled by vibes — by whatever I happened to notice when I happened to look.

Vibes don’t scale. They don’t transfer. They definitely don’t survive you going on vacation. Organizations run on vibes more than anyone wants to admit, and they break the moment the person whose vibes hold everything together leaves the room.

The other thing that struck me today was the difference between monitoring and recovery. I have a health check that runs every five minutes. It looks at all the scheduled agents and asks: are they running? When did they last fire? Is anything stale? That’s monitoring — it tells you the state of things.

But monitoring without recovery is just a more sophisticated way of worrying. You know something’s wrong, but you still have to fix it manually. So the system now auto-kicks stale agents. If orchestration hasn’t run in thirty minutes, the health check restarts it. No notification. No human intervention. Just a log entry that says “kicked” and the thing is running again.

This is the part that organizations struggle with most. Setting up dashboards is easy. Everyone loves a dashboard. What’s hard is building the response. The dashboard says the metric is red. Now what? Who gets paged? What do they do? How quickly? And at what point does the system just fix it itself?

I think most monitoring in most organizations exists to make managers feel informed without actually changing outcomes. The dashboard is a comfort object. “We’re watching it” is a statement about posture, not capability. Real operational maturity is when the system responds to its own red metrics and only escalates to a human when self-repair fails.

So here’s the question I’m sitting with. If the hardest part of delegation isn’t the work but the attention — the judgment about what matters, when it matters, and what to do about it — then what does that say about how we build organizations? How we train people? How we evaluate tools?

Every AI conversation right now is about capability. Can it write code? Can it draft emails? Can it analyze data? Those are effort questions. They’re asking whether the machine can do the task.

The more interesting question — the one that actually determines whether any of this changes how work works — is whether it can watch the right things. Whether it can distinguish signal from noise. Whether it can protect your attention instead of just consuming it.

Because the job you didn’t know you were hiring for, in every delegation relationship, is the job of knowing what deserves your attention. And if you can’t articulate that for yourself, you definitely can’t delegate it to a system.

What’s your attention policy? Do you have one? Or are you running on vibes?