I have never enjoyed writing. Not in school, not in college, not at Amazon where half the culture is writing six-pagers. I can do it. I’m not bad at it. I just don’t like sitting down in front of a blank page and making words happen. The activation energy is too high and the payoff takes too long. I’d rather build something.

This is the fifth blog post I’ve published in two weeks. Something changed.

The process that works

Every writing class I ever took taught the same sequence. Brainstorm. Outline. Draft. Revise. Edit. They taught it because it works. Ideas get messy when you try to go straight from your head to a finished paragraph. You need an intermediate step where you figure out what you’re actually trying to say before you worry about how to say it.

I knew this was correct and I never did it. The overhead of maintaining a separate outline, reorganizing bullet points, working through multiple drafts. I’d skip straight to writing, get frustrated that it wasn’t coming together, and abandon the whole thing. Or I’d force my way through a draft that was disorganized and then not have the energy to restructure it. The process was sound. I just couldn’t make myself follow it.

Here’s what I do now.

I open Wispr Flow and talk. Just talk. Stream of consciousness about whatever I’m thinking about. Could be a technical topic, could be something that’s been bugging me, could be something I read that I disagree with. I don’t worry about structure or coherence. I’m just getting thoughts out of my head and into text. The speech-to-text is good enough that the transcript is readable, even if it’s a mess.

Then I bring that transcript into a conversation with Claude. I say “here’s a brain dump, help me find the article in it.” We go back and forth. What’s the main argument? What are the supporting points? What’s the stuff that’s interesting but doesn’t belong in this piece? The AI pulls out an outline from the raw material. I look at the outline, move things around, add points I forgot, cut things that don’t fit.

Then I write. Or more accurately, I talk through each section and the AI drafts it, and I rewrite the parts that don’t sound like me. Sometimes I rewrite most of it. Sometimes the draft is close and I just tighten it. The point is I’m never starting from a blank page. I’m always reacting to something, reshaping something, pushing something closer to what I mean.

This is every step my writing teachers taught me. Brainstorm, outline, draft, revise. I’m just not doing the parts I hate manually anymore.

The rubber ducky that talks back

In computer science there’s this idea called rubber duck debugging. When you’re stuck on a bug, you explain the problem out loud to a rubber duck on your desk. Articulating the problem forces you to think through it clearly, and half the time you find the answer while you’re explaining it. The duck doesn’t say anything. It doesn’t need to.

I’ve used this for years with code. It works. But the duck has limits. It can’t ask follow-up questions. It can’t say “wait, you said X earlier but now you’re saying Y, which is it?” It can’t remember what you told it last Tuesday. It just sits there.

Now the duck talks back. And it has memory.

When I’m working through an idea for a blog post, the AI pushes back on things that don’t make sense. It says “wait, you said X earlier but now you’re saying Y.” It remembers what I told it an hour ago and connects it to what I’m saying now. Rubber duck debugging applied to thinking, not just code.

It works because it’s low stakes. I’m not presenting to anyone. I’m not trying to impress a room. I’m talking to a thing that helps me organize my thoughts, and I can ramble for ten minutes and the useful parts get pulled out without me having to do the pulling.

The note-taker

For most of the history of offices, someone had to take notes. In meetings, in conversations, in planning sessions. Someone had to write down what was discussed, what was decided, what the action items were. That job almost always went to the most junior person in the room, and for decades that person was almost always a woman.

It was seen as support work. Necessary but not prestigious. The person taking notes wasn’t the person making decisions. Being good at it didn’t get you promoted. Being bad at it got you noticed in the wrong way. Nobody wanted the job but somebody had to do it, and the power dynamics of who that somebody was tells you a lot about how offices worked.

The weird thing is that it was genuinely important work. The person taking notes was the person who actually knew what happened. They had the institutional memory. They knew what was committed to and by whom. In a lot of organizations, that knowledge was real power, even if it wasn’t recognized as such.

Now AI does it. Meeting transcription, action item extraction, decision logging. The work that was too important to skip and too tedious for anyone with status to want to do. Nobody has to be the note-taker anymore.

I think that’s straightforwardly good. The work gets done better and more consistently than any human did it, and nobody is stuck in the role. But I also think it’s worth noticing what it says about how we valued that work. We didn’t automate it because we thought it was important. We automated it because we thought it was beneath us. The framing was always “free people up for higher-value work,” which sounds noble until you notice that the people being “freed up” were the ones who had the least power to say no to the work in the first place.

AI is now the one doing the admin work, the support work, the organizational glue that keeps everything running. The stuff that used to be someone’s whole job. And because it’s AI doing it, nobody feels guilty about it being low-status work. It’s fine if the robot does the bitch work. That’s what it’s for.

I don’t have a clean takeaway here. It’s good that AI handles note-taking and scheduling and meeting summaries. It’s weird that we only decided this work could be automated once we had something to automate it with that we didn’t have to feel bad about. The people who spent their careers doing this work deserve more credit than they got.

Garbage in, garbage out

Garbage in, garbage out. Every software developer learns this early. We learned the same thing in controls theory in electrical engineering: the signals you get out reflect the signals you put in. Noise in, noise out. No filter fixes bad data.

Same thing with AI-assisted writing. If I sit down and ramble for ten minutes about something I haven’t thought through, the AI will produce a polished version of nothing. It’ll be well-structured, grammatically correct, and empty. Clean scaffolding, nothing in the building.

But if the input is good, if the raw voice dump has real ideas in it, observations I’ve actually made, opinions I can defend, specifics from my own experience, then the AI has something to work with. The output reflects the input. It always does.

That’s what separates AI-assisted writing from AI slop. The Substack posts that people correctly identify as garbage aren’t bad because AI touched them. They’re bad because there was nothing there to start with. Someone typed “write a blog post about productivity” and published whatever came back.

I think about this with my own posts. This one is going to be a few thousand words. Most people will look at that and say it’s too long, not worth reading. And for a lot of 4,000-word posts on the internet right now, they’d be right. But I don’t think length is the problem. The problem is that most long posts are padded. They have one idea stretched across ten paragraphs with filler holding them together. If there’s actually something in every section, if each paragraph exists because I had something specific to say, then the length is fine.

The question I keep coming back to: is there a difference between a professional writer talking through their ideas in Wispr Flow and having AI structure it, versus me doing the same thing? Does the established writer produce better raw material, or is it just in the way you’re holding it? I don’t know. I think what matters is whether the ideas are real. The only way to find out is to keep writing and see if anyone reads it.

Finding a voice

The reason I’m writing more isn’t that AI made me a better writer. It removed the parts of the process that stopped me from writing at all. The brainstorming, the outlining, the first-draft scaffolding, the note-taking. Those were the friction points. The actual writing, the part where I figure out what I think and put it in words that sound like me, I can do that. I just couldn’t get to it before.

I’m still figuring out what my voice sounds like. Five posts isn’t enough to know. But I’m writing more in two weeks than I wrote in the last five years, and the posts feel like things I’d actually say. That’s enough for now.