WHAT ARTIFICIAL INTELLIGENCE (AI) IS AND IS NOT

ARTIFICIAL INTELLIGENCE ASSISTANT PROMPTS These two prompts reflect tasks a working harness engineer faces when a new project is proposed. Replace each bracketed placeholder with your own material before use. [START OF PROMPT] I am scoping a new agent project and need to justify, in a short internal memo, why this task requires an agentic approach rather than a simple rule-based script or a single-call model integration. Project description: [PROJECT DESCRIPTION]. Write a one-page memo that (1) classifies the task as narrow-AI-with-agentic-loop versus a simpler alternative, (2) lists the specific signals in the task that justify that classification (multi-step, needs tool use, needs judgment under ambiguity, and similar signals), and (3) flags any part of the task that could be handled more cheaply and reliably by plain deterministic code instead of an agent. Output format: a one-page memo with a highlighted recommendation line at the top. [END OF PROMPT] [START OF PROMPT] Review the following agent capability list for scope creep toward tasks that would require general intelligence rather than narrow AI, which this harness cannot safely guarantee: [PASTE CAPABILITY LIST / FEATURE BACKLOG HERE]. Flag any capability that implicitly assumes open-ended judgment, common-sense reasoning across unrelated domains, or long-horizon autonomous goal-setting beyond what a narrow, tool-using agent loop (the repeating cycle in which an agent thinks, acts, and observes) can reliably deliver. For each flagged item, propose a narrower, more reliable reformulation. Output format: a table with columns Capability, Risk, Narrower Reformulation. [END OF PROMPT]