I'm an experienced backend developer paired with a mobile developer for knowledge sharing. He's genuinely a good person, but during implementation discussions he often stops mid-conversation, types what I said into Claude, and then reads the response back to me as "Claude says…". The issue isn't that he uses AI—I use it regularly too—but that he seems to defer to it instead of engaging with the technical discussion or explaining his own reasoning. I want to understand how he thinks about the implementation, not have a second-hand AI answer. Has anyone dealt with this, and how can I bring it up without making things awkward?
5 Answers
A small amount of mid-conversation searching isn’t automatically bad; people have always looked things up when they didn’t know something. The problem is the interruption and the lack of synthesis. If you already know what you want from the conversation, tell him plainly: “If I wanted Claude’s answer, I could ask it myself. I’m asking because I want your opinion and your reasoning.” If the behavior continues and affects the work, document specific examples and raise the collaboration issue with your lead.
Be direct but frame it around collaboration rather than criticizing his tool use. You could say, “I’m happy to use AI as a second opinion, but in these conversations I’d really like to hear your approach first. Even if you’re unsure, what do you think we should do?” You can also ask, “Does what I’m saying not make sense, or is there something you want to verify?” That gives him a chance to explain whether this is a confidence issue or simply a habit.
This may be less about trusting Claude more than you and more about not trusting himself enough to answer. Some people use an AI response as a shield because it feels safer than being wrong in front of a coworker. Asking “What do you think?” a couple of times, and making it clear that uncertainty is fine, might help him get back into the conversation.
There’s also a practical workplace concern: if he blindly follows an AI suggestion and production breaks, responsibility still belongs to the humans making the decision. AI can miss project context, invent details, or confidently recommend something unsuitable. A developer’s value is judgment—combining documentation, experience, team input, and risk—rather than simply passing model output along.
The useful distinction is whether AI is helping him reason or replacing his reasoning. Using it to research an unfamiliar API, challenge an idea, or check edge cases is reasonable. Pasting your explanation into it and reading the output verbatim during a call adds another layer instead of moving the conversation forward. The person making the decision still needs to understand the context, explain the tradeoffs, and validate the answer.
Exactly. If he wants a second opinion, you can even review the prompt and answer together. That makes it you and him evaluating a tool, rather than him acting as a relay.
That seems like the least awkward approach. It makes clear that you value his judgment instead of treating AI use itself as the problem.