Traditional prompt engineering is no longer the cutting edge; it has become basic “table stakes.” With the release of highly advanced models (such as versions 4.7 and 5.5), AI agents are now roughly 100 times more powerful than they were just months ago. The central theme of this discussion is a necessary paradigm shift: we must stop treating AI as a junior assistant that requires rigid, step-by-step task instructions. Instead, to accomplish complex, heavy knowledge work, we must interact with AI as a “senior partner.”

The Shift to the AI Question Method

The speaker argues that the term “prompting” implies a simple ask-and-receive transaction. As AI agents increasingly manage vast datasets, call tools, and operate over longer periods, this transactional approach falls short. The solution is the “AI Question Method.” This method mirrors how a good manager interacts with a senior team member: by framing the collaboration as a series of thoughtful questions that explore a problem space, rather than just demanding a predefined output.

Three Principles for Managing AI as a Senior Partner

To effectively use the AI Question Method, users should follow three core principles:

• Shape intent with boundaries (The Flashlight Method): Instead of asking overly broad questions, clearly state your thesis or perspective. This acts as the bright center of a flashlight beam, giving the AI clear direction. At the same time, define the edges by explicitly telling the AI what areas to explore and what specific details to exclude from the final output.
• Ask open-ended questions to synthesize complex outcomes: Do not rely solely on rigid evaluations (evals) to determine “what good looks like.” Challenge the AI with difficult, multi-faceted questions that force it to synthesize diverse requirements—such as balancing hardware and software perspectives in a complex business document—allowing it to figure out the best way to weave ideas together.
• Invite the AI to wrestle with data and push back: When providing the AI with a large, diverse set of files (transcripts, data, internal docs), embed your opinions into your queries. Ask the AI to evaluate your thesis against the entirety of the provided data. Give it explicit permission to challenge your assumptions and present a cleaner, more elegant thesis if the data points in a different direction.

Conclusion

The core takeaway is that the words used in prompts matter less than the strategic intent behind them. By shifting your mental model from dictating tasks to a junior assistant to exploring complex questions with a senior partner, you can unlock highly leveraged, creative, and meaningful work with modern AI agents.

Mentoring question

How can you reframe your next complex AI task from a rigid, step-by-step instruction into a strategic question that invites the AI to challenge your assumptions?

Source: https://youtube.com/watch?v=ogTLWGBc3cE&is=QxF21FSN0du8qGLE