1 min lesson
Prompt and context strategy
Work through this case. "A trial team reports the agent's output is 'mediocre.' Before blaming the model, what do you investigate?" Say what you would do and why.
Step 1 of 2
Prompt and context strategythe difference between reliable and noisy output
Most disappointing output traces back to missing context, not a weak model. Coaching engineers to give the agent the right files, the right rules and a clear intent is the skill that turns a frustrated trial user into a daily user.
Coaching the difference: a vague ask vs. a context-rich one
# Weak - no files, no intent, no constraints "fix the checkout bug" # Strong - scoped context, clear intent, explicit guardrails "@checkout.ts @cart.ts the total is wrong when a coupon is applied. Trace where discount is computed, fix it and add a failing test that reproduces the coupon case. Use integer cents (see project rules)."