Human-first defaults for AI systems
Five defaults we set on every deployment before anyone touches a prompt.
- 01Defaults beat policies in practice
- 02Five defaults: explain, refuse, log, escalate, expire
- 03Exceptions get signed, dated, and reviewed
Defaults decide behaviour more than policies do. If a policy says 'a human reviews sensitive outputs' but the default is auto-publish, the policy loses. Set the defaults first.
Ours, in order: explain before act, refuse before guess, log before respond, escalate before auto-close, and expire before retain. Every workflow inherits them unless there is a written exception, signed off, with a review date.
The point is not the specific defaults. The point is that defaults exist, live in code, and get audited. A human-first system is one where doing the right thing is the path of least resistance.
Voice guardrails for AI content
How to encode taste as a rubric so agents don't drift into slop.
Context is the product
Prompts are cheap; the context you feed the model is the moat.
The zag rule
If your competitor could run your prompt and get the same output, you don't have a brand.