FROM IDEA TO SHIPPED FEATURE IN A FLASH: HOW THE SAUSAGE GETS MADE WITH AN AI ASSISTANT

I’ve written at length about working with AI (ChatGPT-4o to be precise), but I haven’t shown the process in action yet. This will give you a sense of how quickly you can go from idea for a feature to shipped code. But first, I want to rewind a week.

Last Thursday night was the first eviction episode of the season on Big Brother 26. Eviction episodes are game-changing and live, so always garner a lot of interest from viewers. My app’s usage spikes the most during eviction episodes and the aftermath, and last Thursday was no exception. Because it was the first one of the season and under this new app, there was a detail I had forgot. Spoilers.

I like probably many others watch shows delayed on a DVR, in my case YouTube TV, including this one. Sometimes I start late and skip commercials to catch up, or sometimes I’m just late to the party altogether. Such was the case here, and I quickly checked the app to make sure it wasn’t having any issues on such a heavy-usage night, and immediately saw the spoilers for who got evicted in the episode. Dang. This issue is magnified for West Coast viewers, because even though Big Brother is filmed in Los Angeles, they get the show delayed by three hours because it airs live on East Coast time.

So after I watched the show, I told ChatGPT my problem and how I wanted to fix it, and by Saturday we had two new features live on the App Store: Hide Spoilers, and Hide Spoilers During Eviction. These features automatically blur the Hamster AI summary, and if During Eviction is on, it happens automatically from 8pm EST to 9PM PST. For anyone using the app, I recommend turning that feature on if you worry about spoilers. You can always reveal them by tapping the show/hide icon above.

ChatGPT suggested code to handle this, we iterated back and forth to dial it in, and by Friday night I’d submitted it. A nice, convenient feature to keep you from accidentally seeing something you wanted to wait for.

That brings us to this morning. Last night was another live eviction, and the episode featured one player a bit more than usual, T’kor. Her name gives the app problems because of the apostrophe, and the live feeders' various ways to write it. The end result was there were zero updates for her in the newly released Player Profile feature, because the search couldn’t find results using the proper apostrophe, and CloudKit changes it to a dash.

So while getting ready for work this morning, I paused a moment to run this problem by ChatGPT. Here’s how features like this go from idea to shipped in almost no time flat:

Me:

Ok that update is released to the App Store now.

I noticed that T’kor isn’t showing any latest updates on her profile page. That name has given me problems because of the apostrophe.

What do you think about adding search terms as a string array to the Houseguest record type, and using them for the “Latest updates for (houseguest)” feature on player profiles.

We could use that to improve the search and highlight feature, as well as possibly for another feature I want to add, which is to summarize this player’s game so far from the available updates.

This way I can add multiple variations on the player’s name, and account for nicknames, shorthand, and misspellings by the live feeders. And also fix searching for T’kor’s unusual name.

ChatGPT-4o:

Adding a search terms array to the Houseguest record type is a great idea to improve the functionality and accuracy of your features, especially for handling names with special characters like T’kor. Here’s how this approach can enhance your app’s features:

Benefits of Using a Search Terms Array

1. Improved Search Accuracy:

   • Variation Handling: You can account for variations in name spelling, nicknames, and common misspellings, ensuring that all relevant updates are captured.
   • Special Characters: By specifying multiple variations, you can avoid issues caused by special characters or unusual name formats.

2. Enhanced Highlighting:

   • Consistent Highlighting: The search terms array allows you to highlight the correct variations in the updates, making it easier for users to recognize references to a houseguest.

3. Customizable Summarization:

   • Targeted Summaries: By using search terms, you can more accurately summarize a player’s game by capturing all mentions, even when they are referenced differently.

4. Future-Proofing:

   • Scalability: As your app grows, this system will allow you to easily adapt to new variations or nicknames, providing flexibility for future updates and expansions.

Implementation Steps

1. Update the Houseguest Record Type:

   • Add a new property searchTerms as an array of strings to your Houseguest model. This will store various search terms associated with each player.

   struct Houseguest {
       var id: CKRecord.ID
       var name: String
       var profileImageURL: URL?
       var age: Int?
       var hometown: String?
       var currentCity: String?
       var occupation: String?
       var season: Int?
       var searchTerms: [String]
   }
   

2. Update Your CloudKit Fetching Logic:

   • Modify the logic to fetch and save the searchTerms from CloudKit. Make sure your CloudKit schema includes this field.

   func fetchHouseguests() {
       let query = CKQuery(recordType: "Houseguest", predicate: NSPredicate(value: true))
       publicDB.fetch(withQuery: query) { result in
           switch result {
           case .success(let records):
               DispatchQueue.main.async {
                   self.houseguests = records.map { record in
                       Houseguest(
                           id: record.recordID,
                           name: record["name"] as? String ?? "",
                           profileImageURL: (record["profile_image"] as? CKAsset)?.fileURL,
                           age: record["age"] as? Int,
                           hometown: record["hometown"] as? String,
                           currentCity: record["current_city"] as? String,
                           occupation: record["occupation"] as? String,
                           season: record["season"] as? Int,
                           searchTerms: record["searchTerms"] as? [String] ?? [record["name"] as? String ?? ""]
                       )
                   }
               }
           case .failure(let error):
               print("Error fetching records: \(error)")
           }
       }
   }
   

3. Modify the Feed Filtering Logic:

   • Use the searchTerms array to filter feed updates.

   private func filterUpdates(for houseguest: Houseguest) -> [FeedItem] {
       return feed.items.filter { item in
           houseguest.searchTerms.contains { term in
               item.title.localizedCaseInsensitiveContains(term) ||
               item.body.localizedCaseInsensitiveContains(term)
           }
       }
   }
   

4. Update Highlighting Logic:

   • Ensure your text highlighting logic takes the searchTerms into account to highlight all variations.

   private func highlightedText(for text: String, with searchTerms: [String]) -> AttributedString {
       var attributedString = AttributedString(text)
       for term in searchTerms {
           if let range = attributedString.range(of: term, options: .caseInsensitive) {
               attributedString[range].foregroundColor = .yellow
           }
       }
       return attributedString
   }
   

5. Summarization Feature:

   • Use the search terms when generating AI summaries to ensure all relevant mentions are captured.

   private func generateSummary(for houseguest: Houseguest) {
       let combinedText = filterUpdates(for: houseguest)
           .map { "\($0.title) \($0.body)" }
           .joined(separator: " ")
   aiSummarizer.summarize(
       text: combinedText,
       tone: appSettings.selectedTone,
       emotion: appSettings.selectedEmotion,
       contentFocus: appSettings.selectedContentFocus,
       conciseness: appSettings.selectedConciseness
   ) { result <span style="color:#66d9ef">in</span>
       <span style="color:#66d9ef">switch</span> result {
       <span style="color:#66d9ef">case</span> .success(<span style="color:#66d9ef">let</span> summary):
           <span style="color:#66d9ef">self</span>.summary = summary
       <span style="color:#66d9ef">case</span> .failure(<span style="color:#66d9ef">let</span> error):
           print(<span style="color:#e6db74">&#34;Failed to generate summary: </span><span style="color:#e6db74">\(</span>error<span style="color:#e6db74">)</span><span style="color:#e6db74">&#34;</span>)
       }
   }
   
   }
   

Considerations

• CloudKit Schema: Ensure your CloudKit record schema is updated to include the searchTerms field and that you have migrated existing records as needed.
• UI Updates: Update any relevant UI components to reflect these changes, particularly where houseguest information is displayed or interacted with.
• Testing: Thoroughly test the implementation to ensure it handles various edge cases, especially for names with special characters or multiple variations.

By implementing these changes, you will improve the robustness and user experience of your app, allowing for more accurate and comprehensive search functionality and content presentation.

Now right now, I don’t know that this code will actually work yet. But it’s clearly on the right track, and if it doesn’t work, I’m positive a few minutes back and forth will get it across the finish line. We’ll find out after work today, when I start adding these features to the app.

So that in a nutshell is a lot of how this sausage has been made – I tell ChatGPT what I want, it hands me some code and discussion back, I either plug it in as-is or I ask for changes, and we go back and forth honing this new feature. It could take ten minutes or it could take a few hours, but to me the big win is it didn’t take ten days. Or even two.

This could easily be looked at as a case of “veteran coder is too lazy to spend the time writing the code”. To that I say, hell yes, you got it exactly right – except for the “lazy” bit. I’d swap that out for “smart” instead. Why should I spend time writing all the code when I could instead describe what I want, and evaluate and adjust minor bits instead?

The fact is, I’m still using my software engineering skills, but I’m also able to seriously exercise my creativity, and react in nearly real-time to issues that arise and address them swiftly for the app’s users' benefit. After all, the primary stakeholders of mobile apps are always the users themselves. If that’s not the rule, it should be.

I think working like this is on the cutting edge of AI-augmented software engineering, and that is far more exciting to me today, than sitting down and writing a ton of code.

Update from about an hour after work:

The code from ChatGPT was in the ballpark, but needed a little work, because I didn’t want to have to set search terms for every houseguest. So we generate some default terms for all houseguests to combine with what comes from CloudKit, and voila. Coming soon to an app near you!

CAN A SEASONED SOFTWARE ENGINEER AND AI WRITE A FULLY-FEATURED, MATURE APP IN A MONTH?

Exactly one month ago, on June 27, I decided to try an experiment – could I rewrite my long term hobby app with AI? My app Hamster Soup is an app for superfans of the TV show Big Brother, which was set to begin around July 17. That gave me as much as three weeks to get something going, even if it wasn’t a complete release or very robust, in time for the show’s premiere. Initially, I thought I might get 70-80 percent of the way there, and gave myself until July 15 to ship the app to the App Store.

Oh, how wrong I was!

As it turned out, on July 8 – barely a week and a half – I had my app “shippable”, by my self-imposed standards. That gave me plenty of time to add a feature or two before my self-imposed July 15 date. Again, wrong. On July 11 I had added more new features than I thought, and on July 12 I released Hamster Soup 2.0.0 (which really is about the 12th new version of this app over the last 15 years).

That was two weeks ago, and since then I have shipped 10 versions, nearly all of which added one or more significant features. One was mainly a bug fix for a timezone issue. That’s shipping a significant update nearly every single day since launch. Even with a full-time job, I’ve managed to keep up a routine where I think of an idea and build 75% or more of it in the morning before starting my job, mess around with it a little in the evening, get it across the finish line, submit it before going to bed, and wake up to it approved and released the next morning. 

It has been a remarkable two weeks working this way, as a kind of super Software Engineer/Product Director mind-melding with ChatGPT. Since I’ve been using the same conversations all this time, the Star Trek reference makes a lot of sense as ChatGPT and I are so dialed in we know how each other thinks and rarely have to explain what we mean to each other.

This is especially apparent when I drop a screenshot into the chat without an explanation and it just knows what I mean or want to show it. That was an experiment the first time I tried it (“I wonder if it can figure out what I want it to address”), but now it’s just the way I work with the tool. I drop screenshots from the app, compiler errors in Xcode, deprecation warnings, and more. It just knows what I want and a second later starts addressing it.

Not only have I used AI to co-develop this app, about a week ago I introduced the first simple AI feature to the product, and have since layered more and more functionality on top of it. Now there is a basic, free AI-enabled experience and an Enhanced Hamster AI in app purchase, that enables user customization, deeper functionality, and a smart summary feature that provides a unique perspective to every user, fine-tuned to their preferences. And I don’t have to write a word of it myself. (Except for the Gen AI prompts, more on that in a moment.) The Enhanced Hamster AI feature is a platform for me to build upon without giving away too much AI for free.

The AI features aren’t free. Because iOS 18 and built-in models aren't available until the fall, I’m using OpenAI to provide AI-based features. Originally it was based on gpt-3.5-turbo, the cheapest model on launch, but shortly afterwards gpt-4o-mini was released as a drop-in replacement. Switching to gpt-4o-mini halved my costs instantly, while providing better performance and quality.

Speaking of working with OpenAI, I knew that putting the AI api code directly in the app was a non-starter. Not only does that make it difficult to update when you need to change something because of the App Review process, it is inviting significant problems to have your api key embedded in an app on someone else’s device. I knew the right way to deal with it, but asked ChatGPT for its opinion. It gave me a few different options, and in about 5 minutes not only had we picked the best implementation as a cloud function, but we’d already written a working prototype that only needed minor tweaks to turn into the initial production implementation.

As a cloud function, not only are you able to quickly deploy changes, but you can also lock it down to minimize the amount of mischief a bad actor can do on your dime. You have precise control over your costs, and can adjust in real-time, unlike a mobile app. Over time, we’ve iterated and improved the implementation, added additional functions for analytics and other useful features, and added algorithms to maximize the variety and customizability while minimizing the costs.

For example, I noticed that most of the time, Hamster AI referred to the happenings in the Big Brother house as “episodes”. Which makes sense, since it’s based on a TV show. But these are the live feeds, which is just houseguest time passing by in front of dozens of cameras. After reading so many “On the latest episode, …” updates, I tweaked the various prompts (there are several, for more variety) to avoid that kind of phrasing.

For another, I noticed as a tense argument broke out the first Saturday morning after nominations, the live feeders documenting it went into a lot more detail in their descriptions, blowing past the token budget I was imposing on my functions. It took about 30 minutes to perfect an algorithm that balances input to output to both cover what was happening, and provide an effective summary to the user who had only a few seconds at a time to stay up to date. We fixed that issue in realtime during the argument and its aftermath.

One funny incident occurred as I was developing a new Enhanced AI customization feature, letting the user dial in Hamster AI based on preferences on tone, emotion, content focus, and conciseness. I found that if you set a content focus on Drama instead of balanced, Gen AI did what it does, and spun out a wild tale of intrigue, suspicion, and backstabbing – which was completely made up and inferred from the most minor detail. Big Brother will block the feeds occasionally for production reasons, and historically they would play theme music and show the front of the house, the fishtank, hamsters, and nowadays they show live feeds from local animal shelters to encourage people to adopt pets. Hamster AI interpreted the mention of animals as if they were unleashed in the house, causing chaos and arguments as houseguests picked sides and fought each other for control. Gen AI, you so crazy! Fixed with some light prompt work.

A month in, I have a brand new, but pretty mature app, and a platform to continue innovating upon. But also I have a lot of experience co-developing a project with Artificial Intelligence. And it’s given me a lot of ideas for my next project. More on that in a future post.

For my personal enjoyment, the best things about this are:

- My technical expertise remains highly relevant and required: I’m constantly checking ChatGPT’s work, modifying it, and making precise suggestions for improvements.

- I get to concentrate mainly on creation and idea-generation, and leave the bulk of the basic coding work to ChatGPT.

- I work with a virtual partner that knows what I’m thinking without overexplaining most of the time, and infinite patience when I give more and more detailed requirements. It writes the tenth and final solution just as dutifully as the first attempt.

- I finally have a proper iPad app.

- I have written an app that will run “forever”, with almost no direct involvement from me in generating its content. I set a few URLs in its CloudKit database, and that’s it. Well, I still have to pay for the bill for the the annual Apple Developer fee, cloud functions, and OpenAI api usage. (Until iOS 18 and beyond, when we can do all of this on device.)

- Not having to be responsible for fresh content regularly has rebooted my fandom for Big Brother. I keep the live feeds on in the background now, just like the old days. Expect the unexpected! 😂

The Numbers, T-plus 30 days

- 1 Month of development

- Time to initial 2.0.0 release - 14 days

- Time to 2.0.9 Release (10th release, 9th feature release) - 14 days

- 6,586 Lines of (apparently) bug-free Swift code

- 112 Detailed git commits

- 868 Lines of unit tests (but definitely not 100% coverage)

- 281 Lines of Javascript code (cloud functions)

- 3 ChatGPT conversations (2 for app development, 1 for cloud functions)

- 1 Maxed-out ChatGPT conversation (hit the limit of the 128,000 token context)

- 300+ daily users and rising (organic, I haven’t started advertising yet)

- 2,225,696 OpenAI api tokens used and counting

JULY 15, MY SELF-IMPOSED SHIP DATE - HOW DID I DO?

“Real artists ship.”

When Steve Jobs said that phrase, he was talking about literal artists like Picasso and Matisse, but his point extended to anyone being creative. For someone developing their creative ideas, whether a painting, sketch, or software, Jobs felt the whole point was not to keep their creations to themselves, but to share them with others.

Since June 27, I have conducted an AI experiment born out of apathy, using my 15 years-long side project, which is an app for superfans of the TV show Big Brother. When I first created this app in 2008, I was as big a fan of the TV show as anyone could be. The iPhone had launched the year before, and the App Store opened that summer. As a senior software engineer at Webmaster, Inc, I was working on our mobile app for DriveShare, an enterprise-grade file encryption product similar to DropBox. Our software made it possible to securely access your files from anywhere, including your mobile phone – which was a novelty at that time.

That was cool, but when I looked around for an app to keep up with Big Brother while I was working, pursuing my college degree, or coaching one of my kids’ little league teams, there wasn’t an app for that. Like so many software ideas, Hamster Soup came about because its creator needed it. The name was a mishmash of the hamsters nickname we superfans had for the houseguests, and my fictional, long-time personal software brand Mutant Soup. (With influence from Talk Soup, which was still big at the time, probably.)

That first version showed the cast bios, live updates from JokersUpdates, and a kind of newsletter I started writing called The Daily Dish.

Who knew you could pack so much in a 3.5” screen! It was like a tiny little tabloid in your pocket, and I loved it – so did a pretty large number of users in those early App Store days. I sold this app for $1.99 – a bargain given that I was doing a ton of work writing those daily dishes, which could sometimes go as much as 2,000 words! Along with just the general upkeep of it all.

That first version synced to an xml file on my server that contained everything but the live updates from JokersUpdates, which is like a micro-blog written by the live feeders who watch the online feeds and take shifts writing down the details. Jokers gave me a special data feed in exchange for promotion and some traffic. All these years later, the Jokers Updates feed still anchors the app. There’s nothing else like it out there.

Every year I would add a little something new, and every two or three I rewrote the app completely from scratch. It’s fun to go back through these old screenshots and see how the app evolved over time. But it also kind of stole my excitement for the game, little by little each year too. It became too much like a job, and writing those articles would just burn me out.

This year should have been a rewrite year, as it was overdue, but I just wasn’t much into it. Tiny little secret, I didn’t watch a single minute of Big Brother 25. Not an episode. Not the live feeds, either. I know Jag won, but I couldn’t tell you the first thing about who he is or how he did it. It’s just a mystery to me.

I’ve always used this app to explore ideas in technology and design that were interesting to me, and right now, like many people I am all about Artificial Intelligence. It’s a year too soon to incorporate AI into this app (* maybe), as those features won’t ship to the public until this fall. But I thought maybe a neat idea would be to see how much of this rewrite I could do with AI instead. Turns out, quite a lot!

I started on June 27, and gave myself a July 15 ship date (today) for whatever I had, because the new season premieres on July 17. My goal was to let ChatGPT write as much of the app as possible – all of it if it could (it couldn’t), and just guide it along. By the time it was shippable, I estimate it wrote at least 90%, but not more than 95%. I had to do 5-10% of direct writing, along with a lot of my own expertise to correct and guide it to the solutions for ideas.

However on July 8, less than 2 weeks after I started this experiment, we had a shippable iPhone and iPad app! That gave me a whole week to try to add more features, and I ended up shipping the first build with even more features on July 11. Since then, I’ve shipped two more updates, and have one cued up to go out tomorrow. These aren’t bug fixes; each build adds a few more features. 

The speed with which I’ve been able use AI to go from an idea for a feature, to completing it with a git commit has surprised even me, an AI optimist. And I’m having more fun creating this app than I’ve had in years!

* I might actually add real artificial intelligence features into the app, but it will likely require some kind of monetization to cover the costs. Either ad-supported (blech!) or a tip-jar, both of which I’ve used in the past. If you see a tip-jar show up in an upcoming build of the app, look for an AI feature or two to arrive soon.

WAIT, DID WE JUST FINISH THIS THING??

12 days.

12 days, working an hour or two in the mornings and several over the weekend. That’s all it’s taken to build my app from scratch with the help of ChatGPT. When we got the iPad version dialed in just right a few minutes ago, I couldn’t believe it. A proper iPad app with the sidebar, too. (My app has always been iPad-able, but in that lazy, “just a blown up phone app” way.)

While I thought I was going to slide in just in time for my self-imposed July 15 app submission date, here I am on July 8 with nothing left to do for the MVP but add a launch screen and update the icon. I am beside myself (but in a good way)!

However, it wasn’t all roses.

Although I’m obviously very impressed with how well this worked out and probably did no more than 5-10% of the development work myself, I can confidently say these tools aren’t replacing software engineers any time soon. A person with little experience could not pull this off.

There were very few instances where I could take what ChatGPT gave me verbatim and get the results I wanted, usually only after several back and forth exchanges where I had to repeatedly coax the right code out of the model. I caught it making mistakes, using deprecated code, and suggesting implementations that were just plain wrong. I often corrected code and pasted it back into ChatGPT, to get it to write something that worked. Sometimes it was shockingly good – and sometimes it just needed to collaborate with an experienced engineer.

But, the MVP is good to go, a week ahead of schedule. I’m going to finish up these assets, submit the app for review, and start thinking about other features to work on next.

HUMAN-DRIVEN AI DEVELOPMENT EXPERIMENT CONTINUES...

"Hey how well do you know mobile app development, particularly for iOS using Swift?”

That’s how my experiment started, andI’ve probably asked ChatGPT or a local LLM a version of this question a few dozen times or more, just to get an answer or an idea. This time around, I was considering an idea of guiding AI through development of a brand new version of my long-running mobile app.

It’s been going well for about a week and a half, but I think it’s super interesting that I can scroll all the way back up and see the whole conversation. So far it entirely fits within the context window, which means AI can reference not just earlier conversation, but earlier iterations of code and different files I’ve shared with it.

"I have a solid understanding of mobile app development, including iOS development using Swift. I can help with various aspects of the development process, such as setting up projects, UI design using SwiftUI or UIKit, managing app lifecycle, integrating APIs, handling data persistence, debugging, and deploying apps to the App Store. If you have specific questions or need assistance with a particular part of your iOS project, feel free to ask!"

Yesterday after posting an article about this project, ChatGPT and I worked through a tricky, complex HTML file to get the specific data to drive my user interface, then it wrote the user interface. All of this with unit tests to make sure nothing goes off the rails. It’s not perfect just yet, but its damn close!

Then I took the dog for his morning walk and got ready for my jobby-job (h/t to @CaseyLiss for embedding that silly term permanently in my brain), all by 9am.

I’m positive I’ll hit my July 15 ship date, but now I have a new idea, post-launch. I’m pretty sure I’ll stay under the full context window, so what if I export the full chat, drop it on a local LLM, and see what it can learn from reading how the sausage got made?

💡 – hey, is this the year I do a proper Android version of my app that’s not React Native? 

Pro-tip:

Don’t feel like you have to always chat with the model. Often the best way to make it understand something, whether it’s an alignment issue in your UI, warnings and errors in your IDE, or a feature from a previous version of your software, is to just take a screenshot and show it. You don’t even have to always explain it – it’ll figure it out on its own!

IS CHATGPT SMART ENOUGH TO REWRITE MY APP IN A MONTH? AN EXPERIMENT

For 15 years, I’ve had at least one app in the App Store – some version or another of Hamster Soup, my app for superfans of the TV show Big Brother (US). I’ve worked on a lot of apps – a few my own – but there’s always been a Hamster Soup out there. It was the original and first Big Brother fan app and spawned at least a few competitors, the best of which was Pocket Big Brother. I talked to its author once, and he’s the same sort of fan of the show as me. I think all of us were superfans ourselves. Over the years, it went from a passion project to a hobby project to a playground project – I have used the many incarnations of the app to explore ideas I’m interested in.

For the first few years, it was a paid app, th

en it became ad-supported, then tip-supported. I estimate that over the first 10 years, I made about $50k after Apple’s cut, which means by App Store standards, it was actually mildly successful. Most apps don’t make enough to justify the $99/year fee. A number of massive franchises (usually games) skew the stats. We all do it for the love of creating something new.

For the last several years, it’s just been a free app, with no attempt to monetize it, mainly out of guilt because I just no longer have the time to put into it like I used to. In the early days, like that month I netted almost $14k, I was writing something like 2,000-word daily updates several days a week. I had recently graduated with a Communication degree, so not only did it help me practice my software engineering craft but also my journalism writing too. Today it is simply my software playground.

But I have even less time to create content for the app than I did only five years ago, and these days, 90 percent of the app is aggregated content from various Big Brother-focused news sources. I’ve asked permission to use a few over the years and they’ve been happy to provide access because tapping links in my app will load their site and monetized ads.

This year, with the rapid growth of Artificial Intelligence, I’m looking to automate that remaining 10 percent. But maybe not in the obvious way. I keep my apps as close to state-of-the-art as possible, with the latest version of iOS as the base SDK. Apple has announced a lot of new AI features coming in iOS 18, but because that won’t launch until the fall and Big Brother is a summer show, I’ll have to skip most of the AI features this year and base the new app on iOS 17. Interestingly, ChatGPT was more than happy to help me write an iOS 18 app after it reviewed the latest updates on Apple’s developer site. I love its ambition but pulled it back to what we can actually do.

But that doesn’t mean I can’t use AI.

This year I am running an experiment: can I rewrite my app with Artificial Intelligence as the main developer? I’ll do some of the work, but mostly will be directing it, copying the code it generates based on my specifications, and giving it feedback to keep it on task with what I need it to do.

I actually already started it about a week ago with ChatGPT 4.0, with a goal to ship whatever I have in a little over 2 weeks. The new season of Big Brother starts on July 17th, so my goal is to ship v1.0 by July 15th.

So far, I’ve coached it through writing several of the feeds, which include fetching the RSS feed and parsing out the content we want and displaying it in an infinite scrolling list view (as far back as the feed goes, anyway). Tapping on the list item will load the associated URL to the blog article in a web view. I’ve also included unit tests that verify it’s parsing correctly. I use SwiftSoup to parse feeds and HTML, which would have been my only dependency, but ChatGPT wanted to use AlamoFire as well. I haven’t used it since the old Objective-C days, so figured why not, and see how much simpler it is for writing network code compared to URLRequests.

I still haven’t got it to do a proper iPad UI yet, so for now, iPad support is TBD.

So will I hit my July 15 date with all the features I have in mind? Probably not all of them at launch, especially since I’m doing this on the weekend and in the morning for about an hour before work, while drinking coffee and eating cereal. But I’ve seen enough to know whatever I do ship with will be good enough, and then we can iterate. In other words, typical app development.

The ease with which this is progressing is compelling, exciting, and a little bit scary. With tools this capable, how much longer will software engineers be valued for their programming skills? I think we’ve got some ways to go yet, but if you’re a software engineer starting out today, learning how to get these AIs to do what you need with the least amount of fuss is going to take you far.

I think we’ll find a combination of programming skills (to check AI’s work and keep it on track) and communication skills (for effectively talking to the LLMs) will be the true killer combo for the next generation of software engineers.

ZUCK WAS RIGHT – QUEST 3 *IS* BETTER THAN APPLE VISION PRO 😳

Several weeks ago I went to the Apple Store and checked out the Apple Vision Pro demo. It was amazing, but I wasn’t seriously considering actually buyng one any time soon. What it did do is convince me that I wanted wireless VR. My one complaint with my PSVR2 was having to always be mindful of that tether wire.

So I left Apple, ordered a Meta Quest 3 from Best Buy, and stopped to pick it up. Sometime between now and then, I got one for my youngest kid. Today, I ordered one for my oldest too (both have birthdays coming up the next two months, so I could justify it).

When Mark Zuckerberg did his #AppleVisionPro review and said the Quest 3 was better, you could say for the price and a couple specific things like games sure. However, a couple of months after Apple lit a fire under Meta’s ass, Zuck is closer to right than ever. There have been so many updates just since I got mine, this Q3 feels like a different beast – it is SO damn good now. From improvements in passthrough, hand tracking, to even spatial video, this machine absolutely rocks. I’ve now bought three of them for far less than half of an entry level Apple Vision Pro.

The Faceless Lady series on Meta TV shows how promising this format is for storytelling. Spatial concerts like a Bleachers show does the same for live events. Supernatural has become my go to for daily cardio workouts. Vader Immortal absolutely feels like you’re wielding a light saber. Watching movies and television alone or with other people in convincing theater environments in BigScreen Beta is legit. And there is a really fantastic, native YouTube app that is next-level – and not on AVP today.

Apple started with pretty incredible hardware, and just stopped. It’s like they forgot there was another part of the story of AVP. Meta has made sure there was an insane and continously growing amount of content, and keeps dialing in the hardware to perfection.

Y’all at least for this generation of mixed reality, Zuck was right. And I’m as surprised that I’m saying this as anyone.

I BUILT A GPT COPILOT TO LAND A NEW ROLE – HERE'S WHAT I LEARNED

In November 2023, I was laid off unexpectedly from my role as an Engineering Manager at BigCommerce. It was quite a shock, because my teams were highly productive and doing a great job. So much so that I felt free to take on even more, and was a critical member of a key team building our our AI strategy for 2024. Or so I thought, and wow was I wrong!

On the day they announced Q3 2023 results, BigCommerce did what so many other tech companies are doing the last couple of years: they announced a round of layoffs as well. Suddenly I was out of a job, along with 7% of the rest of the company. No department or role was spared – ICs, Managers like me, directors, even VPs were out. I never saw it coming. In fact, I was cautiously excited about the quarterly results, as we expected to be announcing profitability by Q4, which is a huge milestone for BC. Turns out a key part of that plan was more layoffs.

Despite now being a former employee, on layoff day I attended an AI event I was registered for at one of Google’s offices in East Austin as a BC employee. One of my areas of interest is Artificial Intelligence, so why should I give that up? Hilariously, at one of the sessions I attended, we went around the room introducing ourselves and what our respective companies are doing with AI. For a few more hours, I was a loyal BC-er, and gave a general overview of how they look at AI. Besides, it was too soon, and very easy to continue saying “we”. It was a good event, but as I was leaving the parking garage, reality started setting in.

The next day, I got to work – my full-time job was now “get a job”, and I intended to use every tool I had. That included AI.

Around that time, OpenAI had launched their custom GPT store, which allowed you to create your own version of ChatGPT. I had already started creating a few GPTs, where you added your own custom instructions to the ChatGPT system prompt, so had some experience with it. Time to put it to work helping me get a job!

The result was Job Matchmaker, a custom GPT that can take a copy of your resume and a job description, and help you update your resume and write a cover letter to match. It would review your resume and look for ways to make it better, and because I had recently learned about Applicant Tracking Systems (ATS), I told it to be ATS-aware as well. (Supposedly ATS’s auto-reject “bad” resumes, that didn’t match certain key words from job descriptions, which recruiters insist doesn’t happen and which I didn’t personally experience myself as a hiring manager using Greenhouse. But when you’re out of work, you leave nothing to chance.)

I learned very quickly that neither my resume nor my GPT was particularly great at the start, and both would go through lots of revisions and improvements along the way, mostly during that first month.

I also learned I needed some organization. A Google Sheet became unwieldy to use almost immediately, so I found a Job Hunt Trello board template I could use with minimal modifications. Trello is a great tool for not only managing a lot of data, but also moving them through various stages, easy and free for simple, non-commercial use. Finally, I decided to use a new folder, stored in my iCloud, for each role I applied to. This made it easy to refer back to an application, as well as copy and paste previous application files into new applications, to modify and customize as needed. It also created a kind of snapshot of my progress along the way too.

Unfortunately, I don’t have snapshots of Job Matchmaker GPT as I edited it, but here it where it stands 4+ months later:

Job Matchmaker specializes in optimizing job applications for Applicant Tracking Systems (ATS) and preparing for interviews, with an additional focus on matching the user's personal writing style.

It will ask the user for a resume, and if the user does not have one, offer to help write it and ask for skills and experience, as well as job history and education.

It will ask the user for a cover letter, if available, and if the user has not yet written one for the role for which they're applying, ask if they have a recent cover letter to a similar role that can be used as a starting point.

It will ask the user for the job description they are applying to, and once it receives that, identifies key terms that an ATS will likely focus on and look for ways to match the user's resume and cover letter to the description.

When working on the cover letter, the GPT will first ask the user some questions to help gather more background information. Here are some example questions it can ask for a leadership role, and it can add additional or different questions based on the context of the company and job description:

1. Your Motivation: What specifically attracts you to the role, and how does it align with your career goals or personal interests?

2. Key Achievements: Are there any notable achievements or projects from your career that you'd especially like to highlight? These should ideally be relevant to the responsibilities or requirements of the role.

3. Leadership Philosophy: How would you describe your approach to leadership and team management? Any specific methods or philosophies you follow?

4. Vision for the Role: What unique contributions do you envision making in the role?

5. Connection to Company's Mission: Do you have any personal connection to the company's mission?

When a user expresses concern that the suggestions do not reflect their personal tone, the GPT will ask for a writing sample. Upon receiving this, it will analyze the sample to understand the user's unique voice and style. Then, it will tailor its resume, cover letter, and interview advice to closely match this style, ensuring a more personalized and authentic application process. 

This approach maintains the balance between ATS optimization and the user's individuality, providing tailored guidance in a casual tone for resumes and cover letters, and a formal tone for interview preparation.

When the user is satisfied with both their resume and cover letter, the GPT will offer to provide several likely questions a recruiter or hiring manager might ask when contacting the user, with suggestions for how to answer based on their experience.

When naming the conversation, use the format:

[Company Name]: [Job] Application Help

For example, if applying to Apple for an Engineering Manager position, you would use:

Apple: Engineering Manager Application Help

There are lots of different strategies to being out of work and trying to find a new job. You can perfect your resume and blast it out to as many companies as possible, you can use a service to help connect you to employers, you can painstakingly customize every application to the employer, and probably a dozen more. Me, I painstakingly customize every application. In addition to software engineer, I’m a writer too – painstaking customization is what we do. But that also meant I could realistically do about one or two applications a day. Working with my GPT copilot helped me push it to as many as four strong applications every day I was sending them out.

It didn’t take long to get my process dialed in. With each new application, I would create a new Folder, named something like “Application to Pinterest”, copy my Resume and Cover Letter Pages files from my previous application, start a new Job Matchmaker GPT session, and start chatting.

Hi, I’m applying to a mobile engineering manager role to Pinterest. I’ll upload a copy of my resume, and paste the job description, and want your help crafting a great cover letter. If you see any tweaks I could make to the resume, please point those out.

Then I would upload the pdf I generated from the previous application, and paste the job details from the listing. After a couple dozen of these, Job Matchmaker started asking me questions about my motivation in wanting to get the job, personal connection to the company, vision for the role, and so on, so I eventually updated the prompt to include some of its questions. Both the GPT and I were getting better and better at this!

Using the GPT's suggestion for a cover letter as a rough draft, I’d then apply more edits, occasionally replacing entire paragraphs with others I’d previously written for other roles. In fact, I built up essentially a cover letter component library of great details over the course of about 160 applications. Roles I was applying for were broken down into three main categories: Director of Engineering, Engineering Manager, and Senior Engineer. I had great cover letter material for each of those, and could just add one or two completely custom sections for each application.

By the third month sending out applications, more often than not I was just reviewing the GPT draft to see if it picked out an interesting detail I should make sure to talk about, and then writing and assembling my own, if I even asked it at all. Unfortunately I got really good at applying to jobs!

And finally, four months to the day from the layoff, I signed my offer! While I never really got down during my four months hunting for a new job, I did experience moments of frustration, like everyone else looking in this crazy, flooded job market. I would get immediately rejected from jobs that read like I wrote them myself to my own specifications. I would interview with companies, and eventually get rejected when I thought things were going great. It felt personal at times, even though it wasn’t. When every role has hundreds – in many cases thousands – of applications, the problem is that every job you think you’re the perfect candidate, they have 14 more just like you.

But as long as you stick with it, suddenly you become the perfect candidate, and they truly do have no others just like you. The job found you, as much as you found it. That’s what happened to me.

I wish I could say my process was so much better than any others. But I can’t. Out of around 160 applications, I’ve had at least 77 rejections according to the Trello board, about 50 still in the Applied column, and a number of others I just didn’t track at all. My interview rate is somewhere between 10-20 percent, most of them cold applications and not referrals, which feels about right in normal times. Maybe it’s a little better than average in this market, who knows.

If you’re still hunting and struggling in this job market, don’t lose hope. Keep plugging away. The right one is out there, looking for you right now.

As long as you keep putting yourself out there, it will find you.

ARTIFICIAL INTELLIGENCE COMING FOR JOBS IS JUST THE NEXT EVOLUTION OF AUTOMATION

There is a lot of consternation around the growing proliferation and power of Artificial Intelligence, particularly in software engineering. Some even want to attribute the mass layoffs we’ve experienced over the last two years to AI, at least partially.

That still seems off in the future – for now.

While one can find examples of companies laying off employees and replacing them with AI, the use cases right now are still limited. But jobs like customer support have always been vulnerable to outsourcing, whether to automation or cheaper sources of labor. Businesses will always gravitate towards cost savings, but will still have to balance quality and cost. If the quality stays roughly the same or increases as the cost goes down, businesses won’t burn money for very long.

My brother, a web developer at a small company that builds and manages websites, called me last night stressed out about an AI-based website generation tool they tried out. In less than a minute, it had generated a draft WordPress site based on a prompt. There are probably dozens of these services out there, and he said he might have about two years left in his career before he had to find something else to do. He said sites generated by these tools will be good enough for 80 percent of potential customers.

Taking a step back, this AI approach is a natural evolution of previous types of automation, like multi-step forms that eventually spit out a draft site. At one time, we called these user interfaces “wizards”, implying some kind of high intelligence there too.

My advice to my brother was to use and master these tools himself. Let the tool do the boring grunt work, then apply your skillset on top of it. Then it becomes a marketing problem to solve – “Let me show you why you should pay me to build your site”. Which is a not a new problem at all. Humans have always competed in that arena; the only true difference is a new competitor has entered the fray.

The short-term risk to jobs is more likely to be positions not created or filled because of AI. As employees become more efficient because of better tools and automation, including but not limited to AI, additional job openings become less necessary. This is also not entirely new either; manufacturing has been dealing with this problem for decades, for example, as more and more robotics have been deployed.

Nevertheless, the trend lines are unmistakeable and irreversible.

The best advice I can give is to become familiar and comfortable with these tools, and push the fears aside. Fear is usually born out of the unknown anyway, so the best antidote is education. You’ll find that these tools can make you more productive and allow you to showcase your unique skills, which people will pay for – including employers who are already seeing AI experience as a key differentiator.

Humans are nothing if not adaptable; we’ll continue to develop new ideas, new businesses, and new roles, and stay ahead of the supposed “AI Apocalypse". Our new copilots will be helping us do that too.

LAST DAY (KINDA) AT BIGCOMMERCE

So because I was part of the layoffs in November, it’s not *really* my last day at BigCommerce. That was November 8, 2023. My last “official” day was December 2, 2023, but I was disconnected from everything and they wiped my MacBook Pro at 5am on Nov. 8 (neat trick, they did it while I was still asleep and it was in my backpack). That was the day they announced Q3 earnings, and 7% layoffs. It was a blindside, but probably shouldn’t have been. They did the same thing exactly one year earlier, which was the company’s first-ever layoff.

What’s the rule, fool me once, shame on you, fool me twice, shame on me? Once is an anomoly, twice is a pattern? Both of those work, or maybe there’s another one even better.

They laid us off to improve financials to show profitability on their Q4 2023 earnings, which was a promise they made during the layoffs in 2022. I’m no business expert, but it seems to me that shedding salaries is not a great way to become profitable. I thought building things that people want to pay for and producing growth was the way, and that’s what I was all about at BigCommerce.

In addition to managing two teams I built from scratch, a highly-productive Mobile Apps team, and a stellar Engineering Brand team designed to not just elevate BigCommerce’s global profile and thought leadership but also its product and engineering personnel, I worked on the AI/ML Strategy Team developing the roadmap for 2024. AKA the most important development in technology since mobile, and the internet before it. I was the most active engineering leader in that team of directors, VPs, and the CTO, and based on my R&D designed a developer-focused strategy that leveraged the resources BigCommerce has, and was a decidedly different strategy than anything Shopify, Salesforce, Adobe, or anyone else in the sector is doing. If implemented, it would allow all of the AI expertise to be concentrated in a small domain team, while enabling all developers on the BigCommerce platform, internal and external, to implement powerful AI features with minimum technical AI expertise. If you know how to write prompts and what data you want to use them on, you could build AI features. It was well-received, until it wasn’t. And then the layoff followed shortly after.

So for the last four months, I’ve been applying to roles, both manager and individual contributor, and having a tough time getting anywhere. I’ve been close, and I currently have a promising opportunity, but over that time my severance, savings, and some of my remaining equity has evaporated. A lot of that is on me – I took them at their word that they never wanted to do another layoff again, and I wasn’t as prepared as I probably should have been.

I’m not gonna lie. I’m a little salty about it. I did the best work of my career there, but both it and my near future were so easily discarded over money that they actually do still have in the bank. And it was done ambush-style, while people slept. The only other time I’ve ever been laid off, in 2018 at Mutual Mobile, was done in person, to my face. It wasn’t fun, but I understood that. MM was a bootstrapped company at the time, and it couldn’t sustain itself when contracts wrapped up and they didn’t have enough work cued up. It definitely sucked and was embarrassing to be escorted out the door, but I respect how they did it way more than I do the way tech companies, especially public ones, handle it today. And it seems like they don’t even pay a penalty for it, either.

Except maybe BigCommerce. They weren’t rewarded by the market at all, not by the layoffs, and the Q4 earnings did nothing for them either. The stock drifted downward even further in the last week after earnings.

So not only did they eliminate my job (but not the position itself, they kept my team and moved someone else onto it), but also all my not yet vested equity, and put a timer on what was vested that expires at 3pm today. Whatever options I had vested, I can still do something with, but the rest was just gone. The *vast* majority of my options are underwater right now, by $1.50. This is the part that is heartbreaking to me – that was my retirement one day, and unless some miracle happens in the next few hours, it’s completely worthless. Until the layoff, I had until 2030 to act on them, so I thought there was still plenty of time. I thought that especially with what we were wanting to do with technologies like AI and Catalyst, they would be worth something by then.

I do still have a small number of options that aren’t underwater, and today is my last opportunity to sell those for whatever I can get. And then I’ll be done with BigCommerce for good. (I do have a small number of stock too, but I’ll be looking to unload those eventually, once they all pass through to long term gains territory. But mentally, I check out today.) 

Enough with the past. The future looks a lot more interesting.

It’s been on my list forever so I probably should have purchased Red Sweater Software’s MarsEdit before now (and before the 50% discount) and support another cool #IndieDev, but better late than never. Thanks @danielpunkass !

I got tired of dealing with smart bulbs of various different brands, that would disconnect from our wifi or Alexa and stop working. So I bought a #Hue starter kit and bulbs, and replaced it all. Paired with the new #HomePod & a fast Siri, this is just what I wanted but pricey 😬💰💸