Train yourself to notice AI speaks with the same linguistic confidence whether 100% correct or entirely wrong — verbal confidence is not an accuracy signal.

R, A, C, I per task.

Place each risk on 2x2: probability × impact — treat high/high first.

When he opens sensitive topic, say "tell me more" vs reply.

Replace the rote "how are you" with "what was the best part of your week?".

Before any project starts, write a one-page charter answering six core questions.

Start from delivery day and reverse-engineer what must be ready each week.

Compress any project into one page: why, what, who, when, success looks like...

For each task and person: R responsible, A accountable, C consulted, I informed.

A living table of every project risk: probability, impact, owner, action.

For each task estimate: optimistic, most-likely, pessimistic — then compute expected.

Rate each risk on a 5×5 grid — color dictates response urgency.

Per task: who Responsible, Accountable, Consulted, Informed — no duplication for first two.

Per key stakeholder: what to send, when, via which channel, what to expect.

Project first meeting: explicit expectations — scope, date, quality, comms, alarms.

In few-shot, deliberate variety (positive/negative/edge case) matters more than count.