I have been using Claude for my personal life. I am yet to use it to automate my life, like what every other person on LinkedIn is boasting about. But I am actively using the Project mode, specifically where a Claude project has been set up as a PM career coach to think through what the next phase of my professional career would be like, be it in my current workplace or another place.
I have also been actively using Gemini at work - Gemini and ChatGPT are the official AI approved at my workplace, in addition to Cursor. The main project I use here is a “Co-Pilot” that acts as a thinking partner, also with explicit instructions to challenge me and be blunt.
What I observed is that the content that I use with Gemini, has actually made me worse and less confident. I don’t see this yet with Claude, but I wondered why. Initially I thought that this difference stems from the fact that I metabolise the insights from Claude and don’t merely consume it. I see myself challenging Claude’s output more, but in a cognitive way. I come out of it more confident than I do with Gemini.
The obvious challenge to this was that maybe Claude is just validating me and Gemini is actually the one challenging me. I don’t think that’s what’s happening, but the reason isn’t simply that I push back on Claude more. When I look at what I do with Gemini, I’m not really challenging it either - I’m dismissing it. The recommendations don’t fit my work reality, so I move on. Dismissing and challenging look similar from the outside, but they leave you with very different things. Challenging produces something you can use. Dismissing produces nothing.
But the part that actually clicked for me is this: Gemini’s outputs don’t survive contact with the real world. My manager, my stakeholders - the recommendations fall apart in the room. That’s the thing that’s been eroding my confidence, and I hadn’t quite named it. When Claude’s outputs survive that contact (and I think they mostly do), it’s because the stress-testing happened earlier, in the conversation, before I walked into a discussion. With Gemini, the stress-testing is happening in the meeting, in front of the people who matter, and the cost of being wrong is much higher there. The cognitive work I tried to skip earlier still has to be paid - just under worse conditions and with an audience.
The location of the friction is the thing. With Claude, I do the work upstream. With Gemini, my stakeholders do the work for me, downstream, and I absorb the cost in real time.
The uncomfortable question this raises is whether the problem is really the tool, or the setup. My original instinct was that this is about Claude vs Gemini. But if I’m honest, the way I use Gemini - quick recommendations, under time pressure - prevents the metabolising step from ever happening. I’m not sure Claude would survive the same setup. Both things could be true. But they have very different implications: if it’s the tool, switch tools; if it’s the setup, no tool will save me, and I need to change how work-AI fits into my day so that the metabolising happens before the meeting, not during it.
I think the honest answer is that it’s mostly the setup, and the tool difference is doing less work than I initially thought.