Start with the problem, then explore widely
Start with a real problem, test a broad idea and refine whatever proves useful.
5 min read
The useful question is not where AI can be added. It is where a normal task feels slow, repetitive or difficult to begin. Start there, give the tool enough room to surprise you, then shape the result with the context and judgment it does not have.
Start with the problem, not the tool
AI is easier to use when it is attached to a real point of friction. That might be a blank page, a pile of information that needs sorting, several directions that need comparing, or a repetitive task that keeps interrupting more valuable work. The problem gives the experiment a reason and a way to judge whether it helped.
Starting with the tool usually produces a search for somewhere to put it. Starting with the problem keeps the work grounded in an ordinary need. If the AI does not make that task clearer, faster or more open to exploration, it has not earned a place in the process.
Take a bigger first swing
Early prompts do not need to be careful miniature versions of the final answer. A broad request can expose directions that would be hard to reach by moving one safe step at a time. Ask for several approaches. Push one idea further than feels sensible. Let the first result create better questions for the second.
This is closer to sketching than delegating. Most sketches are not meant to survive. Their value is that they make an idea visible enough to react to. AI can produce that raw material quickly, which leaves more time for comparison, correction and combination.
Treat the output as material, not an answer
A confident response can still be shallow, incorrect or wrong for the situation. The output becomes useful when it is questioned. What assumption did it make? What context is missing? Which part is genuinely new, and which part only sounds complete?
Refinement is where the work becomes specific. Add the constraint the tool could not see. Replace the generic phrase. Check the factual claim. Keep the unexpected connection if it opens a better direction. The first response creates material. Judgment turns that material into something worth using.
Keep the useful parts of the experiment
Not every experiment needs to become a system. A prompt that helps once can remain a prompt. A rough prototype can answer the question it was built for and stop there. The point is to notice which parts repeatedly remove friction and which only make the process feel more advanced.
Over time, the useful patterns become easier to recognise. AI fits into everyday work when it supports thinking without demanding that every task be rebuilt around it. The tool can change. The habit of starting with a problem, exploring widely and refining carefully remains useful.