AI Can Multiply Your Thinking. First, Bring Some Thinking.
By Om Umrania
At Masters’ Union, one idea kept showing up across classrooms, CXO sessions, career panels, conversations with founders, and one-to-one career counselling: AI multiplies what you bring to it.
Bring a clear problem, useful context, and a point of view, and it can help you think faster and go deeper. Bring a vague prompt and accept the first response, and you can end up with the same assumptions you started with, expressed much more confidently.
That’s the part of the AI conversation I think deserves more attention. We talk about how quickly we can get an answer. We should also ask what happens to our judgment along the way.
Start with a point of view
The easiest workflow is familiar: ask, accept, move on.
It gets something done. But when we use it for a question we’re supposed to understand, we can skip the most valuable part of learning: deciding what we think and why.
Having a point of view doesn’t require being certain. It requires being able to explain three things:
- This is what I currently believe.
- This is why I believe it.
- This is what could prove me wrong.
The third line matters as much as the first two. Without it, a hypothesis can become something we defend simply because it was ours.
This is what I meant when I wrote that being wrong with conviction may be more valuable than being agreeable with artificial intelligence. A view we can examine gives us something to learn from. Passive agreement can leave us with an answer we haven’t really understood.
The conviction has to come with a willingness to change.
Give AI something to challenge
Take a hypothetical product decision. Suppose a team believes that shortening an onboarding flow will help more people complete it.
Asking AI to suggest a shorter flow might produce useful ideas. But it starts from an assumption that hasn’t been tested: the number of steps is the problem.
A more useful starting point would be to state the hypothesis, explain the evidence available, and ask for the strongest argument against it. What else could explain the drop-off? Which assumptions are doing the most work? What evidence would make the team abandon the idea?
Perhaps the problem is unclear language. Perhaps people don’t understand the value of finishing. Those are possibilities to investigate, not findings to accept because a model suggested them.
That distinction is important. Asking for a counterargument is only useful if we then examine it. Otherwise, we’ve just replaced agreement with an equally untested objection.
The next step is to check the evidence and update the hypothesis. The model can help widen the questions. The decision still needs a reason we can explain.
Leave room to revise
The workflow I’m arguing for is simple:
- State the problem and your current hypothesis.
- Give AI the relevant context and evidence.
- Ask for the strongest counterargument and what you might be missing.
- Check the claims and assumptions against the evidence.
- Revise your view when the evidence demands it.
None of this requires getting the first answer right. It does require staying involved after the answer arrives.
If a counterargument changes our view, that’s useful. If it exposes a gap we need to investigate, that’s useful too. Even when we keep our original position, we should be able to explain why it survived the challenge.
For me, that’s the more interesting promise of AI: a way to push our thinking further while keeping responsibility for it.
As answers become easier to generate, independent judgment becomes more valuable. We still have to decide which question matters, what counts as evidence, and when to change our minds.
AI can multiply your thinking.
First, bring some thinking of your own.