AI models are designed to give a complete-sounding answer quickly. To get to a truly great answer, you must challenge their output. Ask "Are you sure this is the best way?" or "What am I not seeing?" to force the AI to perform a deeper, second-level analysis.

The debate over using HTML versus Markdown to communicate with AI agents reveals a deeper shift. The primary job of a knowledge worker is no longer to complete a task, but to create the optimal conditions, context, and scaffolding for an AI agent to perform the work effectively.

A powerful workflow is to explicitly instruct your AI to act as a collaborative thinking partner—asking questions and organizing thoughts—while strictly forbidding it from creating final artifacts. This separates the crucial thinking phase from the generative phase, leading to better outcomes.

For many knowledge work applications of '/Goal,' such as vendor evaluation or candidate screening, an external, objective truth doesn't exist. The user must define the criteria for success by supplying a detailed, testable rubric. The AI's role shifts from finding information to applying the user's specific judgment criteria consistently across a large dataset.

Unlike coding, where context is centralized (IDE, repo) and output is testable, general knowledge work is scattered across apps. AI struggles to synthesize this fragmented context, and it's hard to objectively verify the quality of its output (e.g., a strategy memo), limiting agent effectiveness.

The '/Goal' primitive in AI assistants like Codex is not a bigger prompt but a fundamentally different interaction. It defines a desired end state and success criteria, allowing the AI to loop, self-evaluate, and work autonomously until the 'contract' is fulfilled. This moves beyond the standard back-and-forth chat paradigm.

Instead of basic prompting, use an AI agent's "plan mode" to collaboratively outline a complex task, like writing a strategy doc. This lets you align on structure, sources, and verification steps before generation, yielding far superior results. It's like briefing a junior employee.

Effective use of the '/Goal' feature requires a 'Goldilocks' scope. A goal that's too narrow ('fix this line') prevents the AI from finding root causes in dependencies. A goal that's too broad ('improve the system') makes the success criteria too vague for the AI to verify completion. The sweet spot allows for discovery within well-defined, verifiable boundaries.

To get better results from AI, don't ask for the final output immediately. Instead, prompt the AI to first provide a detailed process. This allows you to review and debug its logic, then instruct it to execute each step for a more accurate outcome.