
Right now Claude Code shows me 25 open conversations. In 24 of them it's my turn.

None of them is hard. The switching is. I come back to a conversation and I don't remember where I was or what I've already seen. So I look at all of it again.

There's a piece of advice I keep seeing for this, and I agree with it completely: have Claude put its output on a web page, a plain index.html you open in the browser, not only into the chat. Structured, with pictures, something you can scan. I use it all the time. Today Claude generated 15 variants of one illustration and built a page with all of them next to each other. I answered with a number, 2.4, and that was the whole decision.

This is what it looks like, the selection I get back. This is the one for this very post, and 1.2 is the picture at the top.

![A web page with five numbered illustration variants for this post, each with a short description under it](/illustrations/posts/102-i-kept-rereading-the-same-paragraphs-sources/variants-page.avif)

I keep improving my flows and systems. I look for the parts that slow me down or take my energy. Writing content is one of them, and writing blog posts is a part of that.

<tldr>I read my blog drafts on a local copy of my site, and since today everything Claude changes in a draft stays yellow until I say it's fine, so I no longer reread the parts I already agreed to.</tldr>

## Six minutes of talking

I write for two blogs, this one and the one for [Edicek](https://edicek.com). A post on either starts the same way. I talk about the topic for four to six minutes into [Wispr Flow](/blog/the-office-was-built-for-typing), without stopping. Claude writes the first draft from that. First it has to go through my older posts, 100 of them on this blog, and through the guidelines for how I write and in what tone.

Then I go do something else.

When I come back and start reading, I have a comment by the second sentence. Sometimes by the first. I say it out loud, keep reading, keep commenting, and get to about the middle of the post.

Claude needs a while to work all of that in, so I leave again. And when I come back, I start from the top. The same paragraphs. After the next round, the same paragraphs again.

## Read it where it will live

One thing made this a lot better. I stopped reading drafts as Markdown files.

Claude starts my site on my own computer and I read the draft on the page where it will be published, with its illustration already in place. I'm not checking a file anymore. I'm reading a post, the way you are right now.

It didn't fix the rereading, though. A rendered page looks finished from top to bottom. Nothing on it says what changed since I was last there and what didn't. So I started from the top again, because I couldn't see what Claude had fixed and what it hadn't.

## Yellow

Today I was reading a draft like that in Safari. I sent three comments in three minutes. With the third one I added a rule: from now on, when you change something in a blog post, give it a yellow background. And keep it yellow until I tell you that part is OK.

The next time I opened the page, three passages were yellow. Those three were all I had to read. <mark class="example">Something like this.</mark>

There's almost nothing behind it. Claude wraps what it changed in one HTML tag, and my site already knew how to show that. The rule sits in the instructions Claude follows whenever it writes for this blog. When I say a passage is fine, the tag goes and the text stays.

So yellow means it changed and I haven't agreed to it yet. No yellow means I've been there and I'm done with it. I don't have to go back.

It's so obvious that it should have been there from the first post.

I think a lot of working with AI looks like this. Small things anybody would think of, and somehow nobody does. I only start fixing how I work once it has cost me enough energy to hurt.


I have 25 conversations open in Claude Code, and after every round of changes I read the same blog draft from the top again. Reading it on a local copy of my site helped. Since today Claude also keeps everything it changed in yellow until I say it is fine.

I kept rereading the same paragraphs


Today I wrote down two tips, and both felt too dumb to post.

The first one: I [moved the shortcut that cancels my dictation](/blog/i-took-escape-away-from-wispr-flow) away from Escape. I kept pressing Escape for something else and losing everything I had just said. The second one: I told Claude to [give a yellow background](/blog/i-kept-rereading-the-same-paragraphs) to every sentence it changes in my blog drafts, so I can see where to look.

A keyboard shortcut and a yellow background. You read that and think, sure, that's obvious, anybody would come up with that.

I didn't. I lost my dictation that way many times before I moved the shortcut. And I had published 99 posts on this blog when the yellow background occurred to me today. The post about the shortcut went out the same evening as my 100th. I never thought I'd write a hundred of them. I thought blogs were dead.

When I started programming in 2005, I was 14 and tutorials were the best thing there was. You wanted to build something, you didn't know how, and somebody walked you through the steps. The how was the hard part.

It isn't anymore. With AI I can build almost anything I can think of. I'd been programming for twenty years and [couldn't build an iOS app](/blog/twenty-years-of-programming-and-i-couldnt-build-an-ios-app-until-now). With Claude I built one. The steps got cheap. What's left is thinking of the thing in the first place, and you can't ask for something that never crossed your mind.

So I think a different kind of content matters now. I don't need another explanation of how to do something. I want to see how somebody else thinks about the thing they're solving. What annoyed them, what they changed, why. It doesn't teach a skill. It shows people that the same problem could be handled differently, and they take it from there with what they already know about their own projects.

I think people already sense this. What they do with it is wrong, though. Skills for Claude Code get shared everywhere right now, and skills for AI in general, and people feel that once they install somebody's design skill, design is taken care of. I don't see it that way at all.

To me it's the same as [highlighting passages in a book](/blog/re-read-any-book-in-20-minutes). I highlight for myself. Sometimes my wife and I read the same book, about relationships or about raising kids. I mark it in yellow, she marks it in green. And you can see that completely different things speak to each of us. In the same book, each of us is looking for something else.

A ready-made skill is like that. It belongs to the one person or the one team that tuned it and tested it, with their needs, their knowledge and the state they were in at that moment. It's not a universal solution. What I can take from it is how they thought about the problem. The rest I have to do myself.

Most readers will shrug at a shortcut and a yellow background. That's fine. But there's always a group it never occurred to, and with these two I was in it myself. For that group it opens a box that was closed until then.

I once wrote that people follow you for [how you think](/blog/you-are-the-niche-stop-searching-for-your-content-category), not for what you talk about. Now I'd say how you think is the thing worth posting.

So I'm posting both, the shortcut and the yellow background. Then I'll wait and see whose box it opens.


Today I wrote down two tips that felt too dumb to post: a keyboard shortcut and a yellow background. Anybody would think of them, and for a long time I did not. Building got cheap with AI, so what people need now is a look at how somebody else thinks about the thing they are solving.

Show how you think


If [Edicek](https://edicek.com) raised money tomorrow, I already know the first two people I'd hire. One for marketing. One for video.

Not a developer. The product is the part I can carry myself. Marketing and video are where one person runs out of hours first.

And neither of them would start from a blank page.

On day one they'd get the brand. The colors, the Figtree font, the logo and the paper cut-out style, all of it public at [edicek.com/brand](https://edicek.com/brand).

They'd get the illustrations as a system, not as a folder of files. Ten kinds of pictures, from a changelog entry to a YouTube thumbnail to an email icon, and each kind has its own written rules.

They'd get the video tool I built with Claude Code. Its newest scene is [17 seconds of the old browser extension turning into the new one](https://www.youtube.com/watch?v=Si6-yB96yoc), and [the post about that launch](/blog/nobody-uses-a-feature-they-dont-know-about) shows how both the pictures and the videos get made.

And they'd get the history. Everything I decided about Edicek is written down in files an AI model can read: 17 guidelines, from how a changelog entry should sound to how a video script is built, 47 commands for the jobs that repeat, and a knowledge file that's almost 2,000 lines long. On their first day they could ask an AI how a changelog entry should read and get my answer, not a generic one.

A good hire would onboard themselves. They'd open the repository in Claude Code and ask what we do, why we do it and what they could take off my shoulders. Then they'd keep asking until everything is clear. If something is missing, they'd ask me and then update the repository, so the answer is there for whoever asks next. Almost everything is already in those files, and I couldn't tell them much more myself.

That also makes it a good test of fit. A missing detail is normal. But if they need me to explain what we do, why, and what they could take off my shoulders, then either I made a mistake and I'm not ready for them yet, or I need someone with more expertise and they're not the right fit.

If they're the real thing, people with a track record at a level I've never reached, I'd let them lead. I'm not going to hire someone that good and then make them follow my checklists.

But I'd want them to build on top of this, not next to it. A new marketer shouldn't spend the first month finding out what Edicek looks and sounds like. That part is answered. Their time should go into the things I can't do alone.

The funny part is why any of this exists. I didn't build it for a future team. I built it because I couldn't afford one. [My last project](/blog/pick-a-project-you-can-keep-alive-alone) stalled because its product came from people I paid and couldn't steer. They were building the system for me, not the other way round. So Edicek had to work with just me. Every job that repeats became a system.

This time the system is already there, and whoever I hire first plugs into it on day one.


If Edicek raised money tomorrow, I'd hire someone for marketing first and video second. Neither would start from zero. The brand, the illustrations, the video tool and every decision behind the product are already written down, because I built all of it when I couldn't afford the people. A good hire would open the repository in Claude Code and onboard themselves.

Your first hire shouldn't start from a blank page


Clavard, one of my projects that's shut down now, started as a favor. My piano teacher kept fixing mistakes in the sheet music I bought, and one day I asked him if he could write his own, just better. It was meant for fun. He said yes.

It turned into an online store for piano sheet music. Classical pieces by Dvořák, Smetana, Chopin or Beethoven, each arranged for several levels, from easy to hard, some of them for four hands. Every level had a recording by Eliška Tkadlčíková, an incredible pianist and one of the rising stars of Czech piano, so you could hear it before you paid $8 for it.

I wrote all of its code. 618 commits between December 2020 and March 2024. The store, the payments, a back office, a tool that pulled the scores out of Flat.io, where they were written.

And I couldn't make a single thing it sold.

<tldr>Clavard didn't die because the money ran out. It died because I couldn't make, judge or steer what it sold, and the people I paid were building the system for me instead of the other way round. Accelerators ask how big your idea is. The question I ask now is whether I can keep a project alive alone, and I only pick the ones where the answer is yes.</tldr>

## I was the orchestrator

My teacher Jiří, a conductor, wrote the arrangements, paid by the hour. Eliška recorded them and checked every score. A branding studio came up with the name, out of more than a hundred options, and the visual identity. That cost me about $7,000.

My job was the code, and keeping everyone else moving. I was a programmer and an orchestrator of other people's work. The worst part was that I couldn't challenge any of it. I knew very little about music theory, and I'd only just started playing the piano myself. I couldn't question their work, their decisions or their opinions. I wasn't from the field.

In the first twelve months I paid about $6,300 just for the arrangements. In the end the store had 48 of them: 18 pieces by 12 composers, each in two or three levels. After the start of 2023, no new ones were written.

It wasn't that the money ran out. I could have paid for more arrangements. But I felt the direction and the quality of the pieces weren't where they had to be to change anything in the market, and I couldn't improve them at the pace I wanted. More money would have gone down a well. I didn't have a system I could give those people. It was the other way round. They were building the system for me. That was the stumbling block.

And there were so many ideas. I'd found a place where we could film beautiful videos of every arrangement. The concept was good. The price was high, but it was worth it, and I believe the market was there and still is. We could have sold printed sheet music, which has been a big hit in bookstores, and not only here in the Czech Republic. We could have made collectible cards of the composers, like the car cards I had as a kid with the engine and the top speed, only here it would be the year they were born and how many pieces they wrote.

But I felt that in this setup I wouldn't get there. I didn't have the kind of money the plan needed, and I didn't know the field well enough to carry it. The code was never the problem.

For another year I kept Clavard alive the only way I could. I published what was already arranged and paid for short videos, some from a creator in Melbourne, a few from Fiverr. Nothing new was coming. My last commit is from March 2024.

For everything that made up the product, the sheet music, the recordings, the plans, the ideas, even the marketing videos, I needed other people. They weren't working for me. I was working for them, and I didn't even realize it.

## The question accelerators don't ask

When you apply to Y Combinator, or any accelerator, everyone wants the big idea. A billion-dollar business. I applied with Edicek last year, and [they rejected me](/blog/my-yc-rejection-full-application).

What almost nobody talks about is what happens when the business stops working. The money doesn't come, or you hit some other wall on the way. At that point the size of the idea doesn't help you. What helps is whether you can keep going on your own. Keep iterating and improving it, keep it running, keep answering support. Make sure it keeps moving and doesn't die, or at least doesn't look like it's dying, so the people who use it, or might buy it, don't start to wonder. The show must go on.

So the question I ask now is simple. If it all goes wrong, can I keep iterating on this alone? Maybe slowly. Maybe only at the minimum quality people will accept. But alone.

## Your goals are (NOT) too small

Some of my friends tell me that. That some of my goals are unnecessarily small.

I don't agree. The people I know with huge goals end up disappointed, because without clear funding up front they can't get there. I'd rather have a goal I can reach with what I have.

## The first one I can carry

Edicek is the first project where I cover all of it. Not just building the product, but the distribution too. I made the brand, everything but the logo. I made the tool I make the animations in. I made the website and the system that generates the illustrations. I record the YouTube videos and write the posts, this one included.

It runs on very little. If everything fell apart tomorrow, I could keep going.

Ten years ago, every part of a project like this was a person I'd have to find and pay. An illustration meant a designer. A product video meant someone with a camera. A brand meant an agency, and for Clavard it literally did.

Today most of it is a tool and a bill. The illustrations come from OpenAI, generated against a brand I wrote down once. The animations come from a video tool I built with Claude Code. When I want to film something, I drive to a studio I rent, where the camera and four lights are already set up, and nobody else has to be there.

I don't have to ask anyone or wait for anyone. When I get an idea for how to make something better, or for a new feature, I build it myself and see whether it works.

Even Edicek started differently. I worked with a designer on the logo and the brand. The logo is the only part of her work I still use; I redid everything else. And I waited weeks for each piece. She moved at maybe 5 percent of the speed I needed to release things the way I wanted.

That's why I think my old projects would get much further today, even with the experience I had back then.

Not Clavard, though. I don't think any tool would have written those arrangements at the level my teacher did. The problem was never the tools. I picked a product whose core I couldn't make myself.

If Edicek gets investment one day, I'll hire. Marketing first, then video. And [they won't start from a blank page](/blog/your-first-hire-shouldnt-start-from-a-blank-page).

I'd still take the money. I just won't build anything again that dies without it.


Clavard, one of my projects that's shut down now, sold piano sheet music. I wrote all 618 commits of its code and couldn't make, judge or steer a single thing it sold. The people I paid were building the system for me, not the other way round. Accelerators ask how big your idea is. I ask whether I can keep it alive alone.

Pick a project you can keep alive alone


On Saturday morning the new [browser extension](https://chromewebstore.google.com/detail/edicek/mgpbhmoefhlnapfidailoifagkmnbfff) for [Edicek](https://edicek.com) was done. Version 1.0. It opens straight on one question, why this page matters to you, and it finally looks like the share sheet on the iPhone instead of a grey form from a different product.

Then I opened its listing in the Chrome Web Store. One screenshot. A one-line description: "Quick add bookmarks and notes to your Edicek workspace". Version 0.2.0 from July. 6 users. A single rating, and it's one star.

That's what "shipped" has looked like in Chrome so far. In Safari the extension comes with the Edicek app for Mac, and the new version arrives with its next update.

I planned three hours to make people see the new one. It took three days.

<tldr>A feature nobody knows about gets a fraction of the use it could. People find out through what you make of it: an animation, a changelog entry, a picture, a post. I planned three hours for that part and it took three days and about ten things I didn't plan. I underestimated how much of it there is. Not anymore, and it was only three days because the systems behind it were ready.</tldr>

## Marketable moments

Every change I ship gets exactly one chance to be seen. I call those marketable moments, and for most of Edicek's history I wasted them.

The changelog on the Edicek website has 37 entries. 32 of them are a still picture, and 5 come with a YouTube video announcement. Even those pictures were a struggle for a long time, until I had a brand I was happy with and could generate every asset the way I wanted, in the time I wanted. But a picture of the new version can't show what changed. It shows the result and leaves the difference to your imagination. On Twitter it gets scrolled past in half a second.

So this time I decided the change itself would be the thing people see.

## Saturday: the plan

The plan was an animation, a changelog entry and a few posts. From ten in the morning to about two in the afternoon, it looked like this.

**The animation.** 17 seconds of the old popup turning into the new one, built in the video tool I made with Claude Code: a web page with CSS transitions, posed and photographed one frame at a time. The browser slides in, a cursor clicks the Edicek icon, the old popup opens, and then it rebuilds itself into the new one in 13 steps, top to bottom. The title gets deleted and retyped. The link field slides out. The note is saved and the popup closes.

<div style="display: flex; justify-content: center;"><video src="/videos/posts/97-nobody-uses-a-feature-they-dont-know-about.mp4" controls></video></div>

Most of those four hours went into this one. The first version was all at once, which was unreadable. Then the popup was too small to read in a changelog, so a camera now pushes in until it fills 80 percent of the frame. Then the background, which I generated 20 times until it stopped competing with the popup.

**A marketing cut.** The same animation, but with the tool's timeline visible underneath. That's the version that goes on Twitter, because the tool is half the story.

**The changelog entry.** Written against the guidelines I keep for every entry: open with a situation you recognise, then the change, no jargon.

**The illustration.** Edicek has a full brand system at [edicek.com/brand](https://edicek.com/brand), so today this is the cheapest part. I tell Claude Code: here's the brand, look at the older changelog entries, generate ten variants with OpenAI. Then I pick. 22 drafts for one image. The hand clicking the button kept coming out with an arm long enough to reach across the whole picture.

**The posts.** One on my personal account, an announcement on the Edicek account, this post, and a version for Reddit.

The Saturday part alone ran past 55 messages in one conversation with Claude Code. By two o'clock I thought I was done.

## Sunday and Monday: everything I didn't plan

**The success state.** Animating the old popup into the new one, I saw that the moment after Save was a static picture. Everything else moved. So I animated it: paper cards fall into a stack while the page saves, and a tick lands once it's saved.

**"Done done."** Then I only had to port it to the iOS share extension. Except iOS also has a loading state, so I folded that into the animation too. Ideally you never see it, only when saving takes too long. It looked great, which meant porting it back into the browser extension again.

**A bug.** A screen recording of the new popup showed it jumping right after Save. I hadn't noticed it before.

**A YouTube video, by accident.** I didn't plan one. A real YouTube video takes me a day and a half. But I put an intro line at the start of the animation, the logo at the end like the outros on the Edicek channel, and music under it. That's a regular YouTube video, the one below. For a change this small it doesn't need a voiceover. The video tool learned to play music for it.

<div style="position: relative; width: 100%; padding-top: 56.25%; margin: 1.5rem 0;"><iframe src="https://www.youtube-nocookie.com/embed/Si6-yB96yoc" title="The new Edicek browser extension" style="position: absolute; inset: 0; width: 100%; height: 100%; border: 0; border-radius: 0.5rem;" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share" allowfullscreen></iframe></div>

**The store.** When I went to publish, I found the Chrome Web Store description was still that one sentence. Rewriting it took very little time, because the iOS App Store description already said everything. It's 2,506 characters now, with a tip to open the extension with Cmd+E. Then new screenshots for the store, and while I was at it, screenshots for the Mac App Store.

That's about ten things I didn't plan.

## Tuesday: launch day

The launch was waiting on two reviews: Google had to approve the Chrome extension, and Apple the new Mac and iOS release. They're separate review processes, and I didn't know when either would come through. So I set the launch for October 5: the changelog entry, an announcement on the Edicek account and a second post about the keyboard shortcut, with its own illustration. Both came through today, so all of it goes out a week early.

This morning I found a typo in the text of the video and in its thumbnail. So both got rendered again, right before the release.

And this post doesn't go out alone. Two more go with it: [one about the sheet music store I couldn't keep alive](/blog/pick-a-project-you-can-keep-alive-alone) and [one about my first hire](/blog/your-first-hire-shouldnt-start-from-a-blank-page). The three of them took about four and a half hours with Claude Code, on Saturday, Monday and today. They came with 44 illustration drafts, 62 tweet drafts I still have to go through, and 3 drafts for Reddit. Those will be my first posts on Reddit ever, and I probably won't get to them today.

## Ready for the match

Three days instead of three hours sounds like a bad estimate. It was. But without the systems behind Edicek, it wouldn't have been three days.

Every one of those unplanned things landed on something that was already there. The video tool, which only needed music. The brand and the illustration system, so a picture is a request, not a project. The iOS texts, so the store description was an edit, not a blank page. The guidelines for how a changelog entry should sound.

I didn't get through it because I rushed. I got through it because I was ready for the match.

That's why I don't rush the foundation, and why I like putting time into things other people skip. The next time I need an animation with text in it, I'm ready. The next time I need music under a video, I'm ready.

The hours you spend fine-tuning things that look silly get divided between the dozens of times you use them later without touching them again. At most you add another layer on top, and it goes on as easily as the layer before it.

## What people see

Building in public looks like a lot of fun. From the outside you see a 17-second animation, a tweet, a picture in a changelog. Isolated outputs. And for each of them you could probably name ten or twenty things you'd do differently, things that would surely work better, or things I missed.

You don't see the three days behind them, the 22 drafts of one illustration, the success animation rebuilt for two platforms, the rent for the studio or the bill from OpenAI. And honestly, I don't blame you.

There's a line I keep reading everywhere: building is easy now, distribution is king. I don't buy it. If you can build a product that people actually like, learning to distribute it is a skill like any other. The other way round is much harder.

But this isn't about which half is harder. It's about which half people find out about. Your users do see the feature, but only once they learn it exists, and that means more work again: a newsletter, an in-app notification, a banner that says what's new. A feature nobody hears about, whether they already use the product or might start, gets a fraction of the use it would get if people knew about it.

The extension took its time to build. Making it visible took three days. Both of them are the product.


The new Edicek browser extension was done on a Saturday morning. Making anyone see it took three days and about ten things I didn't plan, and it was only three days because the systems behind it were ready.

Nobody uses a feature they don't know about


Yesterday I spent an hour on the footer of a website almost nobody scrolls to the bottom of. It's a paper landscape cut into six layers, and each one moves at its own speed as you scroll.

Nobody asked for it. No user is ever going to write in about a missing parallax footer.

<tldr>The last twenty percent, the polish nobody requests, used to be rationed by what I could build myself, so anything needing a designer never got done. Now that part costs an hour instead of weeks, which means the list of details worth doing just got a lot longer.</tldr>

## I already made this argument

Back in February I wrote that [the last 20% is the product](/blog/the-last-20-percent-is-the-product). Empty states, error states, micro-feedback, the small interactions nobody requests and everybody feels.

I spent three days that week on microinteractions in the app. A copy button that slides up only after an answer finishes, a short vibration, the icon flipping to a checkmark and back. There are a hundred or two hundred of those in Edicek now.

I still believe every word of it. But there's something I didn't say, mostly because I hadn't noticed it yet.

## The polish I could afford was the polish I could code

Every one of those hundred interactions had one thing in common. I could build it myself, in code, at midnight, for free.

That quietly decided which details got done. Anything that lived in code was fair game. Anything that needed artwork was not.

So the list of things I wanted and never did kept growing. Illustrated empty states. A proper onboarding sequence with visuals. Animated infographics on the landing page. A footer that isn't just a row of links. None of those were rejected on merit. They were rejected because each one meant a brief, a designer, a budget, and weeks of waiting for something I could look at but not change.

The last twenty percent wasn't rationed by discipline. It was rationed by which half of it I could reach.

## An hour

The footer needed artwork and animation, so under the old rules it was never happening.

It took an hour. I generated the scene, cut it into six transparent layers, and tied each one to the scroll position. The reason that's possible at all is the brand system underneath it, which took months to settle and which I wrote about [when the tooling finally caught up](/blog/chatgpt-image-25-fixed-my-two-complaints).

<div style="display: flex; justify-content: center;"><video src="/videos/posts/96-nobody-asked-for-the-footer.mp4" controls></video></div>

An hour is not free. But an hour is a decision you make on a Tuesday. Weeks and a budget is a decision you defend.

## An hour is still an hour

It wasn't an hour of admiring my own work.

The scene is generated as one image, and each layer is then pulled back out of that same picture. Six separate generations invent six different shapes and nothing lines up.

Then the part I didn't see coming. Every layer has to be the whole sheet of paper, not the piece you can see. In a flat scene the back wave is mostly hidden behind the ones in front of it, so what you get handed is that thin visible strip. Standing still it looks perfect. Then it starts moving and the strip is floating in mid-air with a gap underneath it.

Colors drift too. Isolate a muted green and it comes back as pale mint, so you end up naming the hex in the prompt and checking it against the scene it came from.

## The list is the point

Toast style. Notification style. Empty states. Onboarding. The way an infographic moves on a landing page.

None of those are features. You can ship without every one of them and the product still works, which is exactly why they get cut first. In isolation each one looks like a waste of an afternoon.

Documentation sits on the same list and is the clearest case of it. Everyone wants to build the next thing. Almost nobody wants to go back and write down how the last one works, and fewer still want to keep it true six months later. It ships nothing. It just quietly decides whether anyone can use what you already built.

But nobody meets them in isolation. They arrive stacked, across the ten minutes somebody spends with your product for the first time. My guess is that thirty or forty percent of whether a product lands sits in how those ten minutes feel. I can't prove that number and I won't pretend it's research. I've just never seen anything win on correctness alone.

What changed isn't that the list matters. It always mattered. It's that most of it used to be out of reach, and now it's an afternoon each.

## The second nobody counts

Nobody is going to thank me for the footer.

Somebody will scroll to the bottom, see it move, and think about the product slightly differently for about a second.

That second used to cost weeks.


In February I wrote that the polish is the product. What I left out is that I could only afford the polish I could code myself. Anything needing a designer stayed on the list. That changed.

Nobody asked for the footer


When I work, sometimes I just type. With Claude Code, or with AI in general. But sometimes a muse grabs me and I only dictate. I dictate and dictate like I'm possessed.

And while I talk, I do something else. Take a screenshot, for example.

The screenshot comes out wrong, so I hit Escape. That's the habit. Escape throws away a bad screenshot.

It also cancels the dictation. [Wispr Flow](https://wisprflow.ai/) stops listening, everything I said up to that point is gone, and I can't get it back.

It keeps happening to me. Dictating [changed the way I work](/blog/the-office-was-built-for-typing), and I rarely do one thing at a time. I talk to one window and click around in another, and I'm scattered enough without help. Escape is my "never mind" key, so I press it without thinking. Sooner or later one of those presses lands on the dictation.

So I took Escape away from it. Cancelling a dictation is now `Command + Shift + Escape` on my Mac. Three keys I basically never press.

If you use Wispr Flow or any other dictation tool, and you run several things at once like I do, change that one shortcut. Pick the most awkward combination you can think of. Or just click the small dictation overlay with the mouse when you really want to stop.


I kept losing whole dictations in Wispr Flow because I pressed Escape to throw away a bad screenshot, and Escape cancels the dictation too. Cancel now lives on Command + Shift + Escape, a shortcut I never press by accident.

Petr Homoky.

I took Escape away from Wispr Flow.

Straight to your inbox.

I kept rereading the same paragraphs.

Show how you think.

Your first hire shouldn't start from a blank page.

Pick a project you can keep alive alone.

Nobody uses a feature they don't know about.

Nobody asked for the footer.