Open this lesson in your favourite AI. It'll walk you through the why, explain the demo, and quiz you on the try-it list.
Every product team now feels pressure to 'add AI', and most respond by bolting a chatbot onto a screen nobody asked for. The result is a graveyard of features that demo well, cost real money per call, and get used twice. Before you write a single API call, you need a clear-eyed view of which AI features actually move a metric and which are theater for the board deck. This module is the most valuable in the course precisely because the best AI decision is often 'not here, not yet' — and knowing that saves you months and a chunk of your inference budget.
A quick gut-check rubric: score a candidate feature on whether it removes real user effort, whether a wrong answer is survivable, and whether the value beats the per-call cost. Low scores are the graveyard.
Use these three in order. Each builds on the one before.
In one paragraph, explain like I'm new to it: what makes an AI feature genuinely useful versus a gimmick?
Walk me through how I should evaluate an AI feature idea step by step — what questions to ask before committing engineering time.
Given a product with three candidate AI features, how would you decide which (if any) to build first, and what evidence would change your ranking?
# A 60-second screening rubric for an AI feature idea.
def screen(feature):
score = 0
score += 2 if feature["removes_real_effort"] else 0 # saves the user a tedious job
score += 2 if feature["wrong_is_survivable"] else -2 # a bad answer doesn't burn trust/money
score += 1 if feature["value_beats_cost"] else -1 # value per use > inference cost per use
score += 1 if feature["hard_without_ai"] else -2 # a regex/lookup would NOT do this
verdict = "build" if score >= 4 else "maybe" if score >= 2 else "graveyard"
return {"score": score, "verdict": verdict}
print(screen({"removes_real_effort": True, "wrong_is_survivable": True,
"value_beats_cost": True, "hard_without_ai": True})) # build
print(screen({"removes_real_effort": False, "wrong_is_survivable": False,
"value_beats_cost": False, "hard_without_ai": False})) # graveyardpython3 main.py