3 Comments
User's avatar
Elliot Duff's avatar

Excellent idea: My statistics were,

* Partner — ~60% (dominant mode): writing/policy work, iwiki.au, article series — you originate ideas, redirect on substance, catch overclaims, supply facts Claude can't know

* Verify — ~25%: research-heavy claims (provenance chains, sourcing, ATO dispute) — you check findings before accepting them

* Surrender — small, low-stakes only: mechanical fixes (Mac camera bug, mesh network, car drain) where there's little to verify

* Reject — ~0%: no instances found of discarding Claude's work wholesale or routing around it

Mark A. Bassett's avatar

Great piece, John. I’ve developed the SECURE GenAI Use Framework (https://secureframework.ai) and the AI Inherent Risk Scale (AIIRS https://aiirs.ai) to provide guidance to people on using AI. I’d love to hear your thoughts on these.

Crispy Thompson's avatar

I got Partner: ~65-70%, Verify: ~20-25%, Surrender: ~10%, Reject/Avoid: ~0-2%

Starting to wonder if I should surrender more!