What Breaks When You Automate

2 July 2026 · 2 min read

blog contents
  1. Silent failure is the normal failure
  2. Everything you depend on will change
  3. A runbook, not a person who remembers
  4. Make reruns safe

read

Automation rarely fails with an error. It fails by continuing to run after the world underneath it changed: a field renamed, a rate limit hit, a supplier altering a format without telling anyone. The work is not building it, the work is noticing when it drifts. Anything running unattended needs a way to say it is unwell and a written answer for what to do about it.

Silent failure is the normal failure

The visible failure, where a job stops and someone gets an alert, is the good case. The expensive case is a job that keeps running and quietly processes nothing, or processes the wrong thing.

The defence is to alert on absence as well as on errors. If a process normally handles a few dozen items a day, zero items is an event, not a quiet success.

Everything you depend on will change

Integrations break because the other side moved. A field gets renamed, a response gains a wrapper, a limit gets introduced. None of that arrives as a notification you will read in time.

Assume drift. Validate the shape of what arrives rather than trusting it, and fail loudly when it does not match.

  • A field renamed or removed upstream
  • A new rate limit or quota
  • Credentials expiring quietly
  • A format change in a file somebody else generates

A runbook, not a person who remembers

When something stops at an inconvenient hour, the question is what to do, not who built it. That answer belongs in writing next to the thing itself.

A runbook is short: what this does, what it depends on, how to tell it is healthy, how to restart it safely, and what must never be run twice.

Make reruns safe

The most common recovery is running the job again. If running it twice sends two messages or creates two records, recovery becomes its own incident.

Design so that repeating an operation lands in the same place as doing it once. This single property removes most of the fear from fixing things.

questions

How much monitoring is enough?

Enough to answer one question without opening the code: is this doing its job right now? Usually that is a count over time plus an alert when it drops to zero.

Who should own an automation?

A named person on the business side, not only whoever built it. Ownership means someone notices when the process it supports changes, which is what actually breaks automations.

Should automations be documented like software?

Lighter than that. One page per automation covering purpose, dependencies, health check and recovery beats a formal document nobody opens during an incident.

More on this track

  1. How to Automate Your Business Automate in this order: count where the hours actually go, fix the process before encoding it, integrate the systems that make people retype data, add AI only where the input is messy and the output gets checked, and stop when the next item costs more to automate than it saves. 29 August 2026 · 3 min read
  2. Processes You Should Never Automate Automation multiplies whatever it touches. Point it at a good process and you get leverage, point it at a broken one and you get the same breakage at speed and at scale. Four kinds of work should stay manual: anything nobody has written down, anything that is decided rather than executed, anything rare enough that the rule never stabilises, and anything where being wrong is expensive to reverse. 5 August 2026 · 3 min read
  3. Find the Hours Worth Automating Most automation projects start with a tool somebody liked and work backwards to a problem. Start from the hours instead: for two weeks, record what the team actually repeats, how often, and how long it takes. The list that comes back is always shorter than the list people describe from memory, and the top item is almost never the one that gets complained about most. 21 July 2026 · 2 min read