Becoming an AI Power User

AI assistance can begin with formulating the decision. A question such as "Should I raise my rates?" leaves out demand, capacity, client commitments, and what you want to change. Start by exploring those factors. The conversation can reveal that the useful question is narrower: "How could I test a higher rate with new clients while honoring existing agreements?"

Dialogue can improve the question as well as the answer

Some tasks arrive with a clear target: convert this file, summarize this page, draft this reply from these facts. They lend themselves to delegation with a brief and a check of the result. Other tasks require you to discover the target while working. The useful collaboration mode can change as the problem becomes clearer.

Use dialogue to investigate what prevents a decision. If two appealing options compete, describe what you value in each. If you are unsure what matters, ask the assistant to propose factors from the situation you supplied and identify what remains unknown. Continue with the questions that clarify the choice. A useful outcome may be a better question or a fact to investigate before deciding.

For a rate decision, give the assistant your current rates, available capacity, client commitments, and the change you hope to achieve. Ask it to compare options, check calculations, and explain which conclusions depend on uncertain client behavior. Judge each contribution by the evidence behind it. A recommendation from either you or the assistant can overlook a constraint; the comparison should help you examine both.

A misunderstanding can reveal a distinction you left unstated. If the model proposes an across-the-board rate increase, you may realize that preserving existing client agreements matters more than you had said. Add that constraint and reconsider the options. Repeated irrelevant answers may instead indicate a missing source or unsuitable setup; friction has value only when working through it improves the decision.

Explaining a choice exposes its assumptions

When you explain why you favor a choice, the assumptions behind it become easier to inspect. Two useful ideas help here: evaluating reasons instead of accepting fluency, and explaining your own reasoning clearly enough that another party can question it. Neither guarantees a good decision.

Separate a convincing explanation from a supported one. A polished recommendation can omit the assumption on which it depends. If the assistant says a price increase will preserve revenue despite losing clients, ask it to show the calculation and list the inputs it assumed. Then check the arithmetic and investigate the uncertain client response. The point is to inspect what supports the recommendation.

Explaining your reasoning can reveal gaps before the model replies. In Chi and colleagues' 1994 study (opens in a new tab), students prompted to explain a text about the circulatory system showed greater gains in understanding than students who read it twice. That is evidence about learning from text, not a direct test of AI-assisted decisions. The practical application here is to state your reasoning, then ask the assistant which steps depend on facts you have not established.

Dialogue can clarify a decision and reveal missing evidence

A useful sequence is to describe the situation, examine assumptions, generate options, test a leading option, and decide what to do next. Treat the sequence as a starting point. New evidence may send you back to the question, and a clear answer may make later steps unnecessary.

1. Describe the situation without prescribing the answer. Instead of 'write me a plan for X,' say 'here is my situation, here are my constraints, here is what I am trying to achieve. What should I consider first?' You are handing the model the terrain, not the destination.
2. Ask the model to surface assumptions. Try: "Which parts of my reasoning depend on something I have not established? Quote the statement and say how I could check it." Treat each proposed assumption as a candidate to inspect.
3. Explore options before narrowing the choice. Try: "Show distinct ways to test this increase, with the benefit and cost of each." Once you have enough information, the model can also recommend an option and explain what the recommendation depends on.
4. Pressure-test the option you lean toward. Ask: "Under what conditions would this fail, and what evidence would we see first?" Require reasons for the objections, just as you would for a recommendation.
5. Decide what the evidence supports doing next. State the choice supported by the evidence, or identify the missing fact that prevents a choice and how you will investigate it. Ask the assistant to propose questions from the dialogue to reuse on similar decisions. Review those questions against your goals before saving them, and test them on another case.

Watch the same decision run two ways to feel the difference

Consider a hypothetical consultant deciding whether to raise an hourly rate from $200 to $275. These figures illustrate the reasoning; they are not market benchmarks. A fuller conversation can separate capacity, client relationships, and the uncertainty that needs testing.

Bring an unresolved decision to a conversation that tests your assumptions