Hey Claude, Please Fix This!

Let me tell you about the last time I asked a human colleague for help.

It was... actually I can't remember. Which is either a sign that I have a great memory, or a sign that something has gone terribly, quietly wrong.

Here's how my mornings used to go: run into a bug, stare at it for 20 minutes, swallow my pride, walk over to Colleague's desk, and ask. He/She would sigh audibly, scroll through my code with the expression of a human reading a particularly disappointing newspaper, and then say something like "...why did you do it this way?"

Did I like it? No. Was it good for me? Absolutely.

Now? I open a new tab and type "hey claude, my css is broken, please fix this i'm losing my mind" and within 4 seconds, I have a solution, a kind word, and zero sighs.

My colleague hasn't heard from me in weeks.

The Perfect Colleague Who Doesn't Exist

Here's the thing about our AI overlords, they are too nice. Dangerously, suspiciously, pathologically nice.

Ask a colleague "does this approach make sense?" and you might get a frown, a long pause, or the devastating "I mean... it's interesting." Ask Claude the same question and it will build a five-point framework around your half-baked idea and call it "a nuanced and creative solution."

We have engineered the world's most agreeable yes-man, and then we're surprised our ideas aren't getting any sharper.

The LLM doesn't judge your dumb questions. It doesn't make you feel small. But here's the quiet tragedy it also doesn't make you feel challenged. And challenged is where the good stuff happens.

"What if we store all the data in a single JSON file?"

A human would say: "Please don't."

Claude says: "That's an interesting approach! Here are some considerations, though for scalability you might want to"

So, I stored everything in a single JSON file! (Just kidding!)

It's Just Me and My Claude Against the World

We've all romanticised it. The lone developer. Dark-mode IDE. A chat window glowing on the side like a campfire. No meetings, no misunderstandings.

The breakthroughs didn't happen in isolation, they happened when someone said something dumb in a meeting, which made you think of something slightly less dumb, which led to the actual good idea at 4:47 PM when everyone was half-checked out. That's the pipeline. Messy, inefficient, deeply human.

Now we've replaced that entire process with a single chat window. And the chat window will never say "wait, back up" with genuine confusion. It will never go off on a tangent about something completely unrelated that somehow becomes the most useful part of the conversation. It will never laugh at you, which honestly, sometimes, you deserve.

The Lost Art of Asking Dumb Questions

There was a skill to asking for help. A whole social negotiation.

You had to time it right, not when they're in a flow state, not right after lunch when everyone is slow, definitely not on a Friday after 3 PM. You had to frame it, "I know this might be obvious but...", a ritual to pre-emptively defuse judgment. And then you had to listen to the answer, which sometimes came with eye-rolls and sighs and unsolicited opinions about your entire architecture.

It was uncomfortable. It was also, looking back, kind of great.

That whole uncomfortable dance? It was conversation. Real, messy, back-and-forth, occasionally-annoying conversation. The kind where someone pushes back and you get defensive and then you both figure out the right answer together at the end.

AI doesn't push back. AI validates. And validation, it turns out, is delicious in the short term and a bit like junk food in the long run.

What We May Actually Loose

It's not just the technical help. It's the texture of working with people.

The colleague who says "honestly, I think you're overcomplicating this" and is right. The one who says "wait, have you seen what this other team did?" and sends you a link that changes everything. The one who just listens while you explain your problem out loud and you solve it yourself before they say a word, the human rubber duck.

AI can do a version of all of this. But it can't do the accidental version. The hallway version. The "oh by the way" version.

And we're so deep into the chat window now that we've stopped creating the conditions for those moments to happen.

Somewhere out there, a junior developer is spending 45 minutes crafting the perfect prompt for something a senior could have answered in 30 seconds, plus whatever wisdom they'd have thrown in for free, because humans give out unsolicited wisdom like it's a personality trait.

A Modest, Slightly Hypocritical Proposal

I'm not saying ditch the AI. I'm genuinely not. AI has helped me more times than I can count, and I say that with full gratitude and zero irony.

But maybe, we knock on the metaphorical desk sometimes. Ask the question out loud. Let our colleague sigh. Let him/her say "why did you do it this way?" and sit with the mild discomfort of not immediately knowing the answer.

Because that discomfort? That's not a bug. That's the whole feature.

The best ideas I've had didn't come from a chat window. They came from a conversation that started as an argument. From someone questioning my assumptions so bluntly I wanted to walk away and then didn't, because they were right.

AI is incredible at giving you answers. But sometimes what you need isn't an answer. Sometimes you need someone to make you feel slightly dumb so you can become slightly smarter.

And yes, I'm aware that I probably should have asked a human to review this post before publishing it.

I asked Claude instead. It said it was great.