The 30-minute automation stack audit

Three things broke or got more expensive in the automation world this month: a model with a hard shutdown date, an editor that switched to usage-based pricing, and a workflow platform whose upgrade silently kills community nodes. None of them were secrets — they were all announced. The teams that got hurt were the ones with no habit of looking.

This is the audit we run quarterly on our own stack. It takes about 30 minutes.

1. Inventory what's pinned (10 minutes)

Grep your codebases and workflow exports for model identifiers and image tags. Every hardcoded gpt-4-turbo, every latest tag, every workflow node pinned to a specific version is a decision someone made once and nobody owns now.

The output is a list: what's pinned, where, and whether the pin is deliberate. Deliberate pins are fine — they're the point. Accidental pins are deferred outages.

2. Check the deprecation pages, not the changelogs (5 minutes)

Changelogs tell you what shipped. Deprecation pages tell you what stops working. Every provider you depend on has one — OpenAI, Anthropic, Google all publish shutdown schedules. Put the dates against your inventory from step one.

3. Re-read your pricing pages (10 minutes)

Pricing changes don't email you. Pull the current pricing page for each paid tool in the stack and compare against what you budgeted. Usage-based transitions are the ones that bite: the sticker price stays flat while the meter starts running.

4. Write down the verdicts (5 minutes)

For each tool: keep, migrate, or watch. One line of reasoning each. The list is the artifact — next quarter you diff against it instead of starting from zero.

That's the whole discipline. The information is always public before it costs you money. The audit just makes sure someone reads it.