The daemon: 21 jobs that run whether I show up or not

At 5:55 every morning, a program wakes up and checks whether a different program will be able to do its job at 6:30. If the answer is no, my phone gets a message with instructions for fixing it before I've had coffee. Nobody schedules this. Nobody remembers to do it. It simply happens, the way sunrise happens, except more reliably documented.

That program is one of twenty-one scheduled jobs running on what programmers call a daemon — a background program that's always on, sitting quietly on a machine, doing its work whether anyone is watching or not. Each job fires on a cron schedule, which is just the old Unix word for "a task that runs at a set time whether anyone remembers it or not." Together they're the backbone of everything else in this book. Vector, the chief of staff from the last chapter, is the brain. The daemon is the heartbeat.

A day in the life of the schedule

The best way to understand it is to walk the clock.

5:55 AM — the credentials check. The morning brief needs permission to read my calendar and email, and that permission (an OAuth token, in the jargon — a digital hall pass) can silently expire. So a guard job probes it thirty-five minutes early and hollers if it's dead. 6:18 — a janitor job quietly closes out stale pending decisions so they don't pile up. 6:25 — the stuck-work scanner, which deserves its own paragraph in a minute. 6:30 — the morning brief itself, the flagship, which gets all of Chapter 2.3. 7:00 — a health check confirms the brief actually went out, because a watchdog that only checks whether the daemon is alive misses the mornings it was alive but slept through its alarm. 7:00 again — a market job scores the housing markets we track against Ray Dalio's six bubble measures and alerts only if a threshold trips.

8:00 AM — relationship nudges. The daemon reads our contact list for people I've promised to stay in touch with — subs, lenders, pastors, friends — and surfaces whoever is overdue for a call. There's even a Sunday 9:00 AM run so the nudge arrives before church, which is exactly as effective as it sounds.

Seventeen past every hour, 6 AM to 10 PM — voice notes. I talk into my phone from the truck or a job site; the daemon transcribes each note and routes it where it belongs — task list, contact record, idea inbox — then files the audio away. Noon — the initiative engine picks the single most valuable signal it has accumulated all day and sends exactly one recommendation. One. Not fifty. (Chapter 3.2 is about what fifty notifications a day does to a person.) 10 AM and 2 PM — follow-up nudges for commitments coming due.

6:00 PM — meeting prep. The daemon looks at tomorrow's calendar, and for every meeting with an outside attendee it assembles a brief: who they are, our history, what's open between us. The night before, not the morning of. Also at 6:00, a capture job scans the day's work log for lessons worth keeping and promotes them to the memory system (Chapter 2.4). 9:00 PM — an end-of-day reflection card: today's pain points and open issues, written up in a structured form I can review at my own pace, or more likely once a month aggregated for insights by AI.

Then the weekly rhythm: Friday 4:30 PM, a self-improvement job reviews which of its own recommendations I acted on, which I ignored, and what that says about its judgment. Friday 6:00 PM, a consolidated learning digest. First of the month, a review of recurring patterns that might deserve promotion to standing principles. It is, I'll grant, slightly unnerving to employ software with better review habits than most managers I've met — myself included.

The job I didn't know I needed

The stuck-work scanner exists because of an embarrassing discovery: AI systems — and their owners — love declaring things done that are merely started. A project ships phase one and everyone moves on; phases two through five quietly fossilize. So every morning at 6:25, a job scans for exactly that: work marked in-progress that hasn't moved, proposals awaiting decisions nobody made, features built but never turned on. At 7:15 a companion job forces the issue, surfacing the three oldest stuck items and making me deal with them — act, snooze, or kill.

It's the automated version of walking the job site on Friday looking for the corner where someone stacked material "temporarily" three weeks ago. Every operation has those corners. Mine just gets inspected daily, by something incapable of embarrassment.

What all this replaced

It's worth saying that the daemon didn't start this capable. The first version ran six jobs — a brief, some market alerts, email triage, task nudges, an overnight data sync, a heartbeat — and each did its thing well, but none of them talked to each other. The brief didn't know about the alert that fired at dawn. The nudges didn't know my afternoon was wall-to-wall meetings. It was a reliable assistant when what I needed was a proactive partner. The rebuild taught the jobs to share their signals, and the twenty-one duties in today's manifest grew from there. (The count will be stale by the time you read this. Manifests only grow.)

What got replaced isn't one job title — it's the invisible half of several. The assistant who assembles the morning picture. The me who was supposed to remember to call people back. The prep I'd do for meetings, meaning the prep I'd skip. The transcription of every windshield voice memo. The end-of-week review that every management book insists on and every actual week destroys. None of it was anyone's whole job, which is precisely why none of it got done.

What it costs to run

Here's the part that surprises people: most of these jobs cost nothing per run. They're database queries and arithmetic — the manifest literally marks them "no LLM cost," meaning no AI model is involved at all. The market scoring, the pattern clustering, the reflection cards, the watchdogs: free, minus electricity.

The jobs that do call a paid AI model are individually metered. The initiative engine costs about a penny per run, with a fifty-cent daily budget. A classification job that sorts my captured notes is capped at a dollar a day. And everything lives under the system-wide $25-a-day spending ceiling from Chapter 2.1.