I tracked every "ops thing" I did for two weeks. The result was insulting: 8.5 hours a week on tasks that followed a script I could have written on a napkin.
The worst part? None of it was hard. It was just unowned — nobody had ever sat down and decided which of these tasks a machine should be doing.
The audit (steal this step)
For one week, log every ops task with three fields:
- What — "checked disk usage on prod-2", "renewed the SSL cert", "grepped logs for errors", "wrote the weekly status email"
- Trigger — time (every morning), event (alert fired), or mood (I remembered)
- Verdict — could a script have done this? (Y/N)
My verdict distribution after 40 logged tasks: 31 Y, 9 N.
The 9 N's were genuinely judgment work: capacity planning, incident retros, deciding what to monitor next. Those deserve human hours. The 31 Y's were: I looked at a thing, compared it to a threshold, and wrote a sentence about it.
The automation ladder (in order of ROI)
Don't automate by coolness. Automate by pain × frequency:
- Time-triggered checks (disk, memory, cert expiry, backup success). Cron + 10 lines of shell. Kills ~60% of the manual load alone.
- Log triage. One script: yesterday's ERROR count vs. 7-day baseline, output a delta sentence. You read one line instead of grepping for 20 minutes.
- The reporting layer. The weekly status email — what alerted, what was noise, what was fixed — compiles itself from the check logs. This is the one people call "AI magic" when it's really just disciplined logging.
- Event-triggered remediation. Last, and only for actions you've reviewed once: restart a service, clean a temp dir, rotate a log. Pre-approved manifest, human-auditable, dead-man's-switched.
What it looks like after
My week now: ~45 minutes reviewing an auto-compiled report, plus the genuinely interesting work. The 8.5 hours went to zero not because of a genius system — because 31 boring tasks each got a boring script.
The failure mode to avoid: automating everything at once on day one, with no thresholds tuned and no runbook. That's how "the automation paged the whole team at 3 AM because of a log spike" stories happen. One task at a time, tuned, then the next.
If you want the head start
I packaged my own version of this ladder — the check scripts, the log-triage one-liner, the auto-compiling weekly report template, and the pre-approved remediation manifest — in the Automation Starter Pack ($19, one-time). If your ops discipline is at zero, start with the Ops Starter Kit ($14) — it deploys in an afternoon. The full stack including the AI dispatcher layer is in the Ops Mega Bundle.
No subscriptions. Runs on your box. You keep the scripts either way.
What's your most embarrassing recurring manual task? Mine was hand-checking SSL expiry — for a domain I owned. Confess below, I'm collecting them for a follow-up.
Top comments (0)