Ian Provencher
Podcast

Routines · 10

The notifications that keep me in the loop

My systems run themselves — but they reach me when it matters. Here's what each routine watches, how it reaches my phone, and why it earns an interruption. The loop runs both ways: I can dispatch work back from the same phone.

Reminders

1

Anything I would be annoyed to forget is booked in two places that share nothing, so either still arrives if the other fails.

Reminder due

Push + desktop window

A time I set, once or on repeat

A cloud routine pushes one line to the phone, and a scheduled window on the PC opens with a button for each way I might answer. The phone half still arrives with the PC switched off.

Health & heartbeats

3

The scheduled machinery tells me it ran — and shouts if it didn’t.

Job failed

Push

A scheduled job on the agents server enters a failed state

A cron or ops job errored — the one alert I actually want to interrupt me, carrying enough context to triage straight from the phone.

Daily ops digest

Chat

Once a day on the agents server

A summary of what the agents did and what is waiting on me. It arrives whether or not anything went wrong, so a missing digest is itself the signal that something stalled.

Alerting heartbeat

Push

A weekly timer, whether or not anything went wrong

The routine that watches the other routines. Every alert here is a message that arrives when something breaks — so a broken alert CHANNEL is silent, and silence is exactly what a healthy week looks like. A deliberate weekly ping I expect to see makes a dead channel noticeable by its ABSENCE instead of by the outage it failed to report.

Agents & decisions

3

My two agents work on their own schedule; when they need a human, they reach me. One page tells me each morning what is waiting across all of it.

Morning summary

Web push

Every morning

One notification from my Command Center page: how many faults and the worst of them, how much is waiting on me, the week’s Claude spend against the weekly limit, and how many repos moved — my agents counted in. Encrypted to a key only my phone holds.

Won’t wait for the morning

Web push

A new red fault, a new question from an agent, or a snoozed item due back

Between morning summaries the same page buzzes the phone only for what should not sit until tomorrow, and overnight it stays quiet except for red.

Needs a call

Chat

An agent hits a decision only I should make

A gate in the workflow: the agent pauses and pings with the context, and I approve, redirect, or answer from wherever I am.

Deploys & repo

3

The build pipeline and the repo keep me posted; I can dispatch work back from the same phone.

Deploy result

Push

A site build finishes (green or failed)

Every push auto-deploys — the notification confirms it went live, or flags a failed build before anyone hits a broken page.

Site check

Push + desktop window

Early every morning, on the PC

A window compares this site’s systems pages with what actually runs and offers one Publish button, and a cloud routine pushes its one-line result to the phone. It also names any scheduled job whose last run failed, or a daily one that has stopped starting.

Dispatch from anywhere

GitHub Mobile

I file an issue on my phone

The two-way half: I capture a task or bug from GitHub Mobile and the agents pick it up — the loop runs even when I’m away from the desk.

Information last changed · current · kept by hand