How to Use Al Without Letting It Think for You (my full framework)

Summary

Justin Sung presents a decision framework for using AI language models—such as ChatGPT or Claude—without outsourcing the thinking and judgment people need to retain. AI can save time, but relying on it too heavily or using it for the wrong tasks may weaken skills and make others less confident in your work. The goal is to become AI-enhanced, not AI-dependent.

AI-Use Flowchart

Start by asking:

1. Do I have the knowledge or skills to do this task without an LLM?

Tasks might range from high-level work, such as creating product strategy, to routine execution, such as writing an email or setting up a server.

If no, ask: “Will I need this knowledge to make future decisions?”

• No: Ask whether the consequences of getting the task wrong are substantial.
  • No—“I don’t know, and it doesn’t matter”: It is generally fine to use AI to complete the task. This may apply to a one-off task that is outside your expertise and has little downside if imperfect.
  • Yes—“I don’t know, but it matters”: Don’t rely on AI’s answer as if you can verify it. You may lack both the expertise to spot errors and the knowledge to prompt effectively. Instead:
    • Learn enough to assess or do the work.
    • Hire someone with the right expertise.
    • Change the plan to reduce the risk.
    • If AI must be used, conduct a premortem: assume the plan failed and identify in advance why it might have failed. Discuss the risks and uncertainties openly.
• Yes—I will need this knowledge later: Invest in learning it or bring in an expert. This applies when you own an ongoing responsibility, will need to answer questions about the work, or must revise decisions as new information emerges. Using AI to produce a plan you cannot explain or adapt can leave you accountable for a decision you do not understand.

Before proceeding, clarify the urgency:

• If time allows, learn the skill or involve an expert.
• If the task is urgent, consider hiring expertise or revising the plan to limit risk.
• If you must use AI, be explicit about what you do not know and what could go wrong. Flagging uncertainty can help decision-makers reconsider the approach and can build trust.

If yes, ask: “Is AI equally good at doing this?”

• No—AI is not as good: Do the task yourself. AI is often less reliable for complex, multifactorial work that depends on experience, context, interpretation, or judgment—especially when there is no single clearly correct answer.
• Yes—AI is equally good: Ask whether it saves a significant amount of time.
  • Yes: Using AI is probably reasonable.
  • No: Consider doing the work yourself. The extra effort can maintain your ability to think and execute, and help others trust that the work reflects your own standards.

Central Lessons

• Time saved is not automatically a good outcome. Fast production is not useful if the result is poor or you cannot take responsibility for it.
• Use AI where it genuinely helps with execution, not as a substitute for expertise you need to develop. If a skill matters to your future work or personal strengths, outsourcing it repeatedly may make you less capable over time.
• Trust depends on judgment, not just the final output. Managers and colleagues need to know that someone can identify errors, handle nuance, and adapt a recommendation—not merely generate one. Be transparent about how AI was used and what human review occurred.
• When a decision matters, don’t treat a plausible-sounding AI answer as proof that it is correct. The less you know about a topic, the harder it may be to recognize errors.
• Learning faster is part of the solution. Sung argues that improving your ability to learn new skills can help meet workplace demands without becoming dependent on AI.

In short, the two situations Sung considers most suitable for handing work to an LLM are:

1. You lack the knowledge, will not need it later, and mistakes do not matter much.
2. You can do the task, AI is equally good, and it saves substantial time.

In other situations, learn, do the work yourself, hire expertise, or change the plan. If AI use is unavoidable for a consequential task, perform a premortem.

Speakers and Sources Featured

• Justin Sung — narrator and presenter; he also briefly interjects as “Justin from the future” to add examples.
• Sources and perspectives mentioned by Sung: Research on AI use and thinking (not specifically identified); managers, employers, and senior leaders he says he has worked with; and examples involving ChatGPT and Claude. These are referenced in the video but are not separate speakers.