I've been watching the context engineering discussions explode lately. Everyone's trying to figure out how to hand-craft the perfect context so Claude has everything it could possibly need to go off and do what they want.
But I keep thinking... most of the time I just don't know what context matters at first.
Yesterday, I was trying to figure out how to structure a talk. I could have written:
You are an expert conference speaker. I need a 30-minute talk for developers about AI collaboration. The audience is 200 people, mixed experience levels...
I could have focused on finding the prospectus, call for papers, previous talks at the same conference. I could have found old talks I’ve given that feel relevant and have the right tone.
But I realized that I didn't know if it was really about AI collaboration, developer empowerment or something else entirely.
So instead I said: "Help me think through this conference talk I’m working on. It’s about AI but I’m not sure what I want to say yet."
Claude replied:
What's the conference? Is this a technical audience, business leaders, general tech folks, or something else?
And what's driving the desire to give this talk - is it something you've been thinking about lately, a response to conversations you've been having, or an opportunity that came up that made you think "I should talk about this"?
Those questions got me off to the races. They got me talking about the details, how I feel about the topic, and what’s frustrating me about how people are talking about it.
It was much better than having Claude throw a draft at me and then nitpicking the draft in hopes Claude would magically land on the right thing.
I was talking to my friend Dave Herman about it, and he landed on the same observation. Claude kept jumping to terrible COPPA solutions until he said: "Let's first have a dialog where we can discover a solution together." All of the sudden, Claude went from mashing to doing actual research.
When you say "help me think this through," you're admitting you don't have it figured out. Which means you CAN'T hand-craft the perfect context because you don't know what matters yet.
It's like trying to pack for a trip before choosing the destination.
The context that matters emerges through the conversation. Each exchange reveals what's actually relevant. You say "oh wait, actually..." and those little corrections (the way you phrase things, what you emphasize) all add up to Claude understanding what you’re actually trying to do.
TL;DR “Help me think this through” plus some details about the nuances of the situation makes a huge difference in getting what you need from an AI. Once you start having a conversation, all the little nuances in your thinking come flooding out, totally naturally.
The cool thing is: you’re already an expert in using language to communicate nuances. And when you lean on that expertise, it goes a lot further than all of the one-shot “context engineering” in the world.