In my previous post, I’ve been working out why Gemini at work leaves me worse off while Claude at home leaves me sharper. I think I’ve hit the root cause of this. The “metabolising” occurs more “upstream” with Claude than Gemini, where it happens “downstream” at contact with the stakeholders.
This reminds me of the calculator analogy. Using a calculator won’t get you better at mathematics. But being already good at mathematics will help you get more out of the calculator. AI feels the same way to me, and I am convinced, like many of my peers on LinkedIn, that experience and expertise matter more now than ever. The axiom of Garbage In Garbage Out holds true even now.
However, I do realise that this is not a 1:1 analogy. And the reason is that the consequences of Garbage Out are worse with AI than with a calculator. You can verify the answer of a calculator 11 * 24 is either 264 or not. If you know enough math, you can easily verify it. But AI output is not like that. It sounds more plausible. It has to be evaluated against context that’s not always visible - my stakeholders, my manager’s priorities, the political reality(which itself is subjective, as it’s only my read of the room). It’s a much harder skill and more an art than science.
Another part where it breaks down is that nobody picks up a calculator and thinks that they will get better at mathematics using it. They are consciously outsourcing their ability to perform mathematics, to the calculator. The relationship between the user and the tool is clear. With AI, there is this perception, and much of what you see on LinkedIn, that it is often pitched as a substitute for underlying expertise - “AI will figure out our strategy”, “AI will empathise for the customer in a support call”. That is a real trap and is very dangerous. The people who are most harmed by AI are people who not only lack the expertise, but also don’t realise it. A calculator will simply provide an output. It won’t try to prove / convince that it’s solution is correct.
Which brings me back to where I started with Gemini and Claude. Metabolising matters not because the output is better. The act of challening the tool keeps my own expertise visible to me. The day I stop pushing back is the day I lose the thing that told me the tool was wrong in the first place.
Maybe that’s the actual skill worth building in the world of AI. Not prompting. Not workflow design. Just the discipline of staying close enough to my own thinking to notice when the tool is drifting away from it.