Approval gates beat blind auto-send
Automatic drafting is useful. Automatic publishing is where damage starts. This note walks through simple approval checkpoints for messages, reports, and status updates.
AI Field Notes
Lessons that survive contact with real work
The field notes page collects practical best practices: prompt design, privacy habits, dummy data routines, workflow design, approval gates, local AI setup notes, and the small adjustments that make a system trustworthy.
Building public demos with dummy data that still teaches
Good dummy data preserves the shape of the task without exposing real people. This note covers placeholders, fake histories, and edge-case records worth simulating.
Prompt design works better when the output destination is known
If a prompt is drafting for a board, portal, tracker, or email queue, the structure should match the destination from the start. This reduces cleanup and ambiguity.
When a local model is enough
Short summaries, classification, checklists, and routing notes often do not need a remote premium model. This note outlines when local-first choices are practical.
How to audit a workflow without turning it into theatre
The useful review is simple: inputs, outputs, review points, stored data, and who can change what. Everything else is secondary.
Lessons learned from JSE corkboards and trackers
Small changes in labels, status rules, and handoff text can matter more than a model upgrade. This section records those details while they are still fresh.