posts

> ./featuredFeatured

Title card reading Your AI agent passes the demo, and fails the audit, with a subhead reading the whiteboard test for AU insurance, beside a two-layer diagram of a programme layer wrapping a per-decision layer of six primitives (task fit, action constraints, identity and perms, legibility, monitor and escalate, kill switch), annotated with Australian regulation tags CPS 230, CPS 234, RG 271, DDO, ICA code, and REP 798.

Agents and WorkflowsYour AI agent passes the demo and fails the audit: the whiteboard test for AU insuranceTen agent governance primitives across two layers, mapped to Australian insurance regulation and supervisory expectations. The whiteboard test for AU insurance.~ 17 min

> ./recentRecent

Handwritten title, Most human-in-the-loop is escalation done badly, on a cream notebook page, with a small upper-right vignette of a three-node AI loop interrupted by a deep-rust human bottleneck labelled 'the loop everyone draws'.

Agents and WorkflowsMost 'human-in-the-loop' is escalation done badly~ 14 min

Hero card titled 'RPA already solved this.' handwritten in two lines, with a deep-rust underline beneath 'solved', and a vignette in the upper right: a 2018 RPA-bot notebook page connected by an arrow to a 2026 Agent page rendered in deep rust, captioned 'same disease, new vocabulary'.

Agents and WorkflowsWhy most AI agent programs are repeating RPA's mistakes (and the playbook to steal)~ 11 min