Score the demo before the workflow

Demos are optimized for awe. Workflows are optimized for Tuesday. Those are different jobs.

What changed for how you should work with models this week: before you adopt a shiny AI feature into a real process, score the demo on paper. If it cannot clear a short bar, admire it and leave your workflow alone.

Use four checks. Write the answers. Share them with one other human.

1. Job fit - Which recurring job does this replace or assist? If you cannot name the job in one sentence, you are collecting toys.
2. Failure cost - What breaks if the model is confidently wrong once: reputation, money, safety, or only a draft? Match scrutiny to that cost (Verify).
3. Evidence you can reopen - Can you reproduce the demo result on your data, with secrets redacted? A skeptic would ask for the redo, not the keynote clip.
4. Owner - Who signs off when it ships? If “the model” is the owner, stop.

This is a Boundaries habit, not a ban on new tools. Enthusiasm stays. Unexamined workflow rewrites do not.

You can say it without a signup wall: score first, adopt second. Skip the day you would only post hype. Silence beats a fake “must use this” post.

See also

• Field Guide: Start here
• Field Guide: Verify
• Challenge: Surviving the hype cycle
• Checklist: AI craft principles