Lines & Spaces

When I was a kid, and even when I was a bigger kid still living at home, Mum would pull out this little brown recipe book with dozens of handwritten recipes she had collected. It had an orange cooking pot on the cover, with the text “Rezepte”, which is German for recipes. My most recent memories of the book are of it being worse for wear, as you might imagine after at least 40 years of use (Mum says at least 50 years).

At the start of last year, for Mum’s birthday, I had the idea to surprise her with a website containing all of the recipes in the book. It was a good chance to see if I could vibe-code an entire website. Things were much different in January 2025. To vibe-code using ChatGPT, I had to copy and paste code from the app into my coding software. It took some effort, but I put the skeleton of the site together. Now I just needed the recipes.

But I immediately had a problem. I didn’t live with my parents and had no access to the book. I enlisted my Dad. He took a photo of every page.

I fed ChatGPT pages upon pages of images, which it did a decent job of pulling the text out of. Decent, but not perfect. There were errors everywhere, so each recipe needed to be verified. For this, I enlisted my two brothers.

I uploaded the images Dad took alongside the appropriate recipe page so that we could verify each recipe was correct. In most cases it wasn’t, and we certainly didn’t get around to verifying all 178 recipes. However, Mum’s birthday rolled around, and I was able to hand over a mostly complete, if slightly janky, website containing all her recipes.

But the story doesn’t end there. AI has come a long way since January 2025, with apps like Claude Code and Codex. It was time to go back and properly finish the website.

A year and a half later, I asked AI to go back through all 178 recipes and verify that they matched what was in the images, then completely redesign the site. It now has a ‘family login’, so we can continue to verify the recipes and add new ones too. It recategorised all the recipes into more obvious categories and added a ‘cooking mode’ for easy iPad use, with a single step on screen at a time and big buttons for messy fingers.

I registered the domain rezepte.casa. It’s a mixture of German (Rezepte) and Italian, Spanish or Portuguese (casa). I like to think of the translation as “Recipes from home”.

Permalink

OpenAI announced GPT-Live earlier today, a new voice model that powers ChatGPT’s voice mode.

We’re launching GPT‑Live, a new generation of voice models that make talking with AI feel much more like having a real conversation.

I watched all of the introductory videos on OpenAI’s YouTube channel, and something peculiar struck me. In each one, the human would continually cut off and interrupt ‘Chat’ (their name for the AI voice). Also from their newsroom post:

Talking with ChatGPT should now feel much more like a real conversation. You can interrupt with a question, pause to gather your thoughts, or ask ChatGPT to slow down.

I felt deeply uncomfortable watching the videos, in the same way that you might feel uneasy if a friend kept talking over you. In my experience, conversational ‘flow’ means leaving space for others to speak, and finding the right moments to speak yourself. OpenAI, on the other hand, are marketing ‘flow’ as interruption.

It’s concerning that OpenAI seems to be normalising habits that are clearly dismissive, and quite frankly, rude.

To test this feature out myself, I wrote this post with GPT-Live on my phone next to me. It’s remarkable, and certainly changes the way we interact with our AI chatbots. The conversations indeed are more natural, and I was able to speak to ‘Chat’ without feeling any need to jump in and interrupt.

That experience left me with a concern. If conversations with AI are starting to feel human, the way we speak to AI matters a lot. Even if it doesn’t feel rude with an AI, getting used to cutting in could still shape how we speak with others.

If you haven’t tried it yet, open the Chat-GPT app and test it for yourself.

Permalink

From a post called The Next Big Thing from my old blog, The Talk Locker, on June 9, 2016:

There are likely to be bumps along the way, but to return to my original point, these assistants are the next “big thing”, and the interface is conversation. Whether that be voice or typing, chatting to a digital assistant will be how you get things done. It’s artificial intelligence that drives an autonomous vehicle. It’s artificial intelligence that filters your gMail. These assistants are getting smarter all the time, and before too long, I think you will be using them for many day to day tasks to alleviate your time to do more important things. I can already hear people complaining that this will make people more antisocial, to the point where people will have these assistants speak on their behalf to their friends. This is a story for another blog post, but I will just say this: be responsible with this technology. It has the power to bring you closer with family and friends. Use Siri to help plan your day, use Google Assistant to book your next trip. Use Cortana to order a pizza! But call your Mum. Text your friends. Go outside. Use this technology responsibly.

Even I’m surprised I got so much of this right, looking 10 years into the future. Apple, Google and Microsoft still have their assistants, though Siri is the only familiar name. The others have re-branded several times (remember Bard?). Most of these assistants now run on large language models, like ChatGPT. Siri, despite being the first major voice assistant to market, has been the last one to finally be upgraded to a LLM powered digital assistant, as Apple announced this week at their Worldwide Developer Conference (WWDC).

Without my use of Codex, OpenAI's coding agent, I would never have been able to build Passing Notes. As I wrote about in my last post, I’ve tried on and off to learn different programming languages with various levels of success. Even now, I would love to be able to write Swift. But I was able to prompt Codex to build Passing Notes, refine it, and publish to the App Store in about two weeks.

Let’s tackle the most obvious criticism of using AI for almost anything - "it’s taking work away from app developers". In a way, yes, it absolutely is. People will use AI to build apps that they would otherwise have hired someone to do. Companies will hire fewer software developers because the developers that they have on staff can be more productive with AI. But in an important other way, if you took Codex away from me - I could not and would not have made this app.

As I mentioned earlier, I don’t have the coding ability to build the app myself. I also don’t have the money to pay a developer to build the app for me. What I do have is a good sense of how good apps work, and what they should look and feel like. I acted as a product manager with Codex as my app developer. I am still pinching myself that I have an app that I made in the App Store. And I can tell you I’m not stopping. I have at least half a dozen projects in various stages of development - apps, websites, the site you’re reading now - that would only have been a fleeting thought a few years ago.

Prior to GarageBand, if a band wanted to get a track recorded, it would cost thousands of dollars in studio time, mixing and mastering. Now, musicians are able to record their own demos to send to producers, and in some cases even record, mix and master a track themselves and make the charts!

Spreadsheet software, the printing press, digital cameras - this is not a new story. Some people are excited by it, others are furious. My thoughts are the same as they were 10 years ago (almost to the day) - disruptive technologies, such as the new wave of LLM powered agents, have the power to advance or destroy industries in very real ways. Our governments absolutely should regulate against the more destructive uses, but there’s no denying that these personified robots can help us be a lot more productive.

I’ve made some apps just for me - a time tracking app, and a personal finance app. I don’t ever plan to release these apps - but I built them to make me more productive. LLM powered coding agents are democratising software development. People will start building widgets, menubar apps, shortcuts and Safari extensions, to suit exactly their workflows.

I predict (and let’s revisit in another 10 years shall we?) that there will be more software written, by more people, for more niche categories.

On an episode of ATP last month, Marco Arment described how he built an app, Saunter Map, for a charity walk he participated in that can:

• Track your location and progress along the route
• See how far the next restrooms and refreshments are
• Estimate your time of completion based on your current location and target pace

Even for Marco, an experienced app developer, this app would not have been made had it not been for AI - it would have taken too long to build for how niche the app is.

Ten years ago I wrote about "the next big thing" being digital assistants. Well, here we are. Whilst many joked about Google suggesting that people should put glue on their pizza, or ChatGPT confidently telling users that the word 'strawberry' contains two R's, it wasn't hard to see where this tech would end up. What has surprised me is the speed at which these underlying models have improved.

I am excited to see how far I will go down this path that I had only ever dipped my toe into the water of before. Already, I have used Codex to help rapidly improve Pocket Tutor, a web platform for music tutors; I’ve made an iPad app, Passing Notes; I’ve made a website for jazz musicians to look up the best recordings of jazz standards; and I’ve used it to redesign my personal website, as well as creating Lines & Spaces.

Permalink

This evening, at 7:36pm, whilst I was sitting on the couch getting ready to tackle some last minute admin before heading to bed, I noticed an email sitting in my inbox from iTunes Store. My gut reaction was that it was spam, but upon closer inspection, I realised that it was an Apple-generated invoice on my behalf for sales of Passing Notes on the App Store.

It’s the first time I’ve ever received money for software I’ve built. And that’s a big deal for me.

Ever since I was a kid, I was into computers and the web. I used GeoCities to build fan pages for TV shows I was into at the time.¹ This was my introduction to HTML, and I still have the muscle memory for basic HTML thanks to these sites. During high school, I became heavily interested in video production. It helped that Dad was a cameraman, and I had the cameras, lighting and computers at home at my disposal. I used all sorts of weird video editing software in the early days, before eventually settling on Final Cut Pro. I even tried to learn Objective-C on multiple occasions so that I could write iPhone apps, but never pursued it long enough for it to stick.

Without a doubt, these were all hobbies. I ended up making my living from playing, writing and teaching music. Ironically, music is the hobby for many people, including many in the tech world. Everything changed when I suggested to my good friend, Nic Jeffries, that we build an app to help peripatetic music teachers manage scheduling and invoicing. After receiving our first quote of $250,000, I suggested to Jeffo that I try to build something myself.

Enter the no-code movement.

Using a tech stack of Zapier for logic, Airtable as the database, Stripe for payments, and Webflow for UI, I built a web app for music tutors that would charge a parent’s payment method when a music lesson finished, in the same way that your card gets charged as soon as you step out of an Uber. There are many parallels between making software and making music. When you create something that works, be it hearing your composition performed that you’ve poured hours into, or seeing your app do what it was supposed to do for the first time, both are equally addictive and emotional moments.

The advent of no-code tools meant I was able to dip my toes in the water of web and app development in a way that was previously restrictive. Almost nothing remains of the original Pocket Tutor now, as it has gone through two more substantial changes to its core. Whilst the no-code movement opened the door for aspiring developers, nothing could have prepared us for LLMs. I went from asking ChatGPT for help with writing code snippets for Pocket Tutor to entrusting Codex to write entire apps for me.

Everyone has an opinion on AI. Some are vehemently opposed to even the mention of it, others are having it complete their homework. Whichever side you’re on, there isn’t any escaping it, and like any invention that has global reach and impact (I’m looking at you, iPhone), you, the user, can use the tool responsibly or irresponsibly, for good or ill. I have chosen to use it to bridge two of my worlds.

Passing Notes was my first attempt at an iOS app. Part of making it was to see whether I could. Making a website is one thing - anyone can publish something to the web. But to get an app onto the App Store used to mean knowing how to code yourself, or spending tens of thousands of dollars to have someone that knew how to code build it for you. The last few weeks have been a wild collection of ‘firsts’:

• I built an iPad app!
• Apple approved my iPad app, which meant I had an app in the App Store!
• People paid actual money to buy my app!

And then we get to this evening’s ‘first’.

• Apple paid me for my app.

Now, admittedly, Apple paid me about $25. I’m not about to go out and buy a Jony Ive designed Ferrari anytime soon. But that’s about 12 people that spent money on an app I built. Receiving this email has sparked something. Perhaps this could be more than just a hobby.

I tell my students all the time that I don’t expect them to go on to become professional musicians, that’s not why they play an instrument in the school band. One percent of those students might, but the rest will go on to become lawyers, accountants, physiotherapists — occupations that musicians will rely on to defend them, balance their books, and correct their bodies.

And do you know what else musicians rely on? Software. A lot of software. Here’s a typical week for me:

• Apple Music to show a student a recording
• Tempo and Tunable as my metronome and tuner
• Pocket Tutor to schedule and get paid by my students
• ForScore to read sheet music
• ScoreSort to manage my PDF library of band music²
• Passing Notes to quickly write and share musical ideas with students

There’s more! Email, file sharing, gig promotion, band program management, accounting. And then there’s the software that is yet to be built. And that’s where I hope I come in. As the saying goes,

“Choose a job you love, and you will never have to work a day in your life.”

Well, music and computers are two of my happy places. I wonder what happens if I combine two jobs I love?

1. I have since tried to find them but they are lost to the web. back
2. Ben is doing fantastic work in the field of ‘software for musicians’. You should absolutely explore his work. back

Permalink