I had a Big Idea. A rethinking of how my team approached a core part of the product. I believed in it. I started writing a design doc, got through the problem statement in about an hour, and then life moved on. The idea sat in a rough sketch that I’d reference in conversations but never flesh out. Writing a proper design doc meant a week-long detour I couldn’t justify, so the idea just quietly died.
I think most ideas die this way. You have the thought, you feel the energy, and then the cost of expressing it formally enough for other people to evaluate kills the momentum. Meeting notes pile up. Rough sketches collect dust. The bottleneck was never the thinking. It was the gap between having an idea and being able to show it to someone.
That gap barely exists for me anymore.
Something I didn’t plan
A colleague was working on a problem I cared about. They seemed a little stuck, and I had an idea for a direction. I said, mind if I put something together? I gathered references I’d been collecting, rough thoughts from our conversations, a few tweaks to what we’d been discussing, and fed it all into my writing loop with Claude Code. Two hours later I had a sketch design with diagrams that I felt genuinely good about. I had accidentally written a design doc.
Before, that would have been a Slack message saying “I have some ideas, let me think about it.” The ideas would’ve sat in my head for a week or two until I carved out time to write them down properly. By then, the moment might have passed. They might have gone a different direction. The energy might have faded.
Instead, we had something to react to that afternoon.
Three iterations in a week
That same week, I had a design review slot booked for a doc that was already finished, but it wasn’t going to be contentious or particularly interesting for the group. I had another idea, though. A more open-ended design with a genuinely controversial direction, the kind of thing that needed critical feedback from experienced engineers.
Two hours before the review, I messaged the principal engineer and said I was switching topics. I put together the design in that window. We had a fascinating review where I learned about a significant simplifier I’d missed completely.
The next day, I met with those same folks and we found a way to handle one aspect of the design much more safely than I’d planned. I put together another iteration of the doc. Then the next morning I woke up with a batch of ideas for how to get benefit from it even sooner, so I notes-apped my way through them and folded them in that afternoon.
Three iterations of a design in one week. Each version sharper than the last because real humans gave real feedback on a real artifact. Before AI tools, that whole cycle would have been meeting notes waiting for me to eventually find the time to sit down and craft.
The bar didn’t drop
I think that bar comes from my Amazon days, where I’d rewrite design docs until I was sick of reading them. Smart, well-meaning people would pick apart every sentence. I wanted each word to earn its place in the doc. Clearing the field, arranging my thoughts as clearly as possible, meant I’d collect more useful feedback on the actual design and waste less time answering questions that better editing would have prevented.
I still hold that bar. I have a rule that AI can draft or edit, but never both without me in the middle. If I skip that step, it’s slop. I read my drafts aloud. I narrate changes with voice-to-text as I go through the doc, partly to catch what sounds off and partly to capture what I want to change in real time. I’ve built a writing skill that codifies my taste and catches my bad habits before anyone else has to. When it breaks a rule, I update the skill.
The effort didn’t shrink. I spend time at a different altitude in the writing process, lifted out of the low-level rearranging and into higher-order concerns. Does this have an arc? Does it sound like me? What do I want someone to take away? I went from rewriting sentences to directing the whole piece.
Along the way, I developed an eye for what reads like AI. I spot it in other people’s work now. I think everyone is building this sense, and that purely AI-written prose will stand out more over time, not less.
Ideas that survive
The real change isn’t writing speed. It’s that my ideas survive long enough to be tested.
Before, I might have three or four good ideas in a quarter and only manage to formalize one or two of them. The rest lived as rough sketches and meeting notes and half-finished problem statements. Now I can take an idea from a conversation to a sketch design in an afternoon. I can swap a design review on two hours’ notice because the cost of putting a new idea on paper dropped that dramatically.
Every doc I share still costs someone their attention. A design doc on a principal engineer’s desk competes with everything else demanding their time. I take that seriously. By the time something reaches a reviewer, it has survived my editing loop. Their job is to challenge the ideas, not fix the prose. I spend the effort earlier so they don’t have to.
But the thing I keep coming back to: ideas that would have died now get to live long enough for someone to tell me they’re wrong, or that there’s a simpler way, or that they’re exactly right. That feedback is the whole point. I just couldn’t get to it fast enough before.