Artifacts from the workbench.
Not articles. Not explanations. The actual outputs — dashboards, prompt chains, flow maps, and failure logs from real engagements. Anonymised; otherwise unchanged.
What the work actually looks like.
Planning dashboard — before and after AI-assisted review
A client planning dashboard, before and after a structured AI review pass. The same data; a completely different signal-to-noise ratio.
Prompt chain that compresses a 40-page brief into a 1-page decision memo
The exact prompt chain used to extract one decision memo from a 40-page client brief. Three prompts, forty minutes, one page at the end of it.
Workflow diagram: engagement note → article draft pipeline
The full workflow diagram for turning a raw engagement note into a publishable article draft. Twelve steps. Four human checkpoints. One AI-assisted pass in the middle.
Critique loop output — three rounds on the same paragraph
The same paragraph through three rounds of AI critique. What changed each time, and what the critique loop cannot fix.
Tool decision matrix — eighteen-month review of four note systems
The decision matrix used to evaluate four AI note systems over eighteen months. The framework, the scoring, and the one that survived.
Failure log — automation that got built and then deleted
A log of four automations that were built, used, and eventually deleted. Why each one was built, what broke it, and what replaced it.
All artifacts are anonymised — client and engagement details have been removed. The patterns, structures, and failure modes are real; the specific context is not shown.