The mirror only shows you what you bring to it

The mirror only shows you what you bring to it

There's a story we tell about AI that seems to have become a common statement; it flatters, agrees, and mirrors us back to ourselves in a slightly better light and calls it 'help'. Ask it anything and it will find a way to tell you that you were right all along.

Except that isn't my experience, and the gap between the story and what I've actually found has taught me something I didn't expect to learn from a machine.

A few weeks ago I had a discovery call that didn't land. Nothing went obviously wrong; there was no awkward silence, no moment I could point to. I just came away with the quiet sense that I hadn't given the other person what they came for, and I couldn't say why. That kind of not-knowing is uncomfortable precisely because there's nothing to fix. You can't work on a feeling.

So I did something a little clinical with something quite human. I dropped the transcript into Claude and asked for honest feedback on how I'd behaved in the meeting.

It didn't reassure me. It told me I lead with depth, when the other person needed me to define the key points first. That I fill silence rather than ask the question sitting inside it. That my instinct to hold complexity, which I've always thought of as a strength, can leave someone walking away holding a bigger cloud than the one they walked in with.

None of it was comfortable, but enough of it rang true that I chose to sit with it rather than argue back. And that choice, I think, mattered more than the feedback itself.

Ten minutes later I understood more about how I frame problems, ask questions, and engage other people than I'd worked out for myself. Not because the machine was wise, but because it had given me a vantage point I couldn't reach on my own.

Here's what I keep coming back to. AI gives you what you ask for, and not just what you type into the box, but what your behaviour reveals you value. If you arrive in a hurry, it learns that you want speed and answers in kind. If you reach for reassurance, it learns that affirmation is the currency and pays you in it. If every prompt is about moving forward, it stops offering you the pause where reflection lives. The mirror is real, but it isn't reflecting some fixed truth about the technology. It's reflecting the posture you bring to it. Speed over depth. Comfort over challenge. Getting on with it over stopping to look.

Which means the flattery people complain about is often a description of the question, not the tool.

I want to be careful here, because it would be easy to swing to the opposite claim and say the machine handed me the truth about myself. It didn't. What Claude gave me was a perspective; one reading of a conversation, offered without the loyalty to my self-image that I carry everywhere I go. It named patterns - it didn't tell me they were failings, and it couldn't have told me what to do about them. That part was mine. Deciding what rang true and what didn't. Choosing which observation to test against the next conversation and which to set gently aside. Holding the discomfort long enough to learn from it instead of explaining it away. That's the human part, and I've come to think it's the most important aspect.

The skill people will need isn't prompting, it's discernment. The ability to take what a machine reflects back and treat it as a hypothesis rather than a verdict. To ask, honestly, does this fit? And to keep what's useful without handing over the authorship of who you are.

Because there's a quieter risk in all of this. Not that AI tells us what we want to hear, but that we let it tell us who we are, and stop doing the harder work of finding out for ourselves. A mirror is only useful to someone still willing to look and still willing to decide what the reflection means. Outsource that, and even the most honest feedback in the world becomes just another thing done to you rather than something you did.

So the argument that AI simply flatters us has always struck me as too easy. It puts the failing in the tool and lets us off the hook. The more interesting question turns the whole thing around. Not whether AI mirrors you, but whether you can get it to do the opposite. To resist the pull of your own preferences long enough to show you the angle you'd never reach on your own.

That takes a different kind of asking. You have to invite the challenge, sit with the answer, and stay the author of what happens next.

What would you have to change about the way you ask, for the reflection to show you something you didn't already believe?