Big updates this week.

A new app hits the App Store, finally!

And a new community hits Skool.

Plus a new AI model.

Topics:

• Wave is Live!

• AI Coding Evolution

• Grok 4

• Automatic Shorts Part 3

• Learn Automation and AI

Wave is Live!

It finally happened.

After 7 months of vibe coding, Wave has made it to the App Store.

It’s been an amazing learning experience, my first app to the App Store, and really crazy to see what AI can do and how far it’s come in such a short time.

I would not have an app anything close to Wave if I was coding this on my own.

AI has made a massive difference, a true 10x maybe even 100x from my normal output factoring in everything that I would have had to learn, write, and debug on my own.

Of course, Wave is also far from perfect.

So many things to fix, add, and improve, but overall, it’s in a good state.

And like I’ve talked about before, it’s way more important to get it into the hands of users.

We could easily spend years “perfecting” every aspect of the app, but it could also be a massive waste of time if no one likes it or wants it.

Better to see what sticks and risk releasing a less-than-perfect app.

This is a quiet launch for now, though.

No marketing yet, it’s just live on the App Store.

Also, I’m planning to get Android up to date next week, so iPhone only for now.

Marketing will probably start next week, so here’s your chance to be an early adopter.

Let me know if you end up downloading and checking it out, I’ll find some way to mark your account in the database for VIP founder status 😉

If you want to use the proper channels, you can email [email protected] with any input or issues, but you can always let me know directly [email protected] 

Thank you for following the Wave journey, I really appreciate it!

I’m excited to see what happens with it next, and mentally preparing to handle a sudden spike of users in the near future…

Time to ride the Wave.

AI Coding Evolution

I worked on Wave, or Pulse when it first started, at a particularly interesting point in time with AI.

AI models jumped considerably in the last few months and recently, fully autonomous agent coders have been taking off, so the landscape has changed drastically, and my development process changed with it.

When I started, Claude Sonnet 3.5 did most of the work because it was integrated into Cursor better than other models at the time, and it did fantastically with the visual styling of the app, and with laying a foundation for how the app worked.

Most of the overall style is originally from Sonnet 3.5.

But Sonnet 3.5 would struggle quite a bit with more advanced logic and state management, or even with staying on task - constantly changing random values that weren’t in scope of the original prompt.

I would offload more complex things to GPT o1 and o1-pro but it would take time to copy all the code that I needed GPT to look at, paste it into the browser, prompt what I needed to happen, read the response, then apply those change back into Cursor.

Thankfully pretty early on, I stumbled across RepoPrompt thanks to an X.com post from McKay Wrigley which helped this process quite a bit:

Repo Prompt made it much easier to quickly select a bunch of files and format them to easily paste into an AI chat window.

And not long after that, Grok 3 ended up taking the lead as the best model for writing code, so I started using Grok to handle more advanced features like authorization context and realtime chat to work when Claude or even GPT sometimes was struggling to produce working results.

Grok wasn’t perfect, but helped make a lot of progress, and not long after that Gemini Pro came along and proved to be even better at coding than Grok, with GPT o3 not far behind Gemini.

Testing both o3 and Gemini, I found I liked the answers from Gemini a little better at the time, plus Gemini allowed for a huge context window of up to 1 million tokens, meaning I could carelessly dump huge chunks of code in the chat to ask Gemini to update or fix whatever the current issue was, so Gemini slowly became my primary coder.

And that became my main coding process.

Use Repo Prompt to grab the files for context, if not all the files in the project, drop them into Gemini, give it a little description of what I was doing or what I needed, and let it generate an answer.

I would read the full answer first to make sure it was something close to what I was looking for, but I became quite comfortable with copying the entire answer and pasting it back into Cursor chat, telling it to apply these suggested changes.

In the end I did end up switching to use GPT 4.1 in Cursor chat specifically because it’s better at understanding step by step instruction.

Most of the time 4.1 could digest a large answer from Gemini and apply all the changes correctly, and even fix any issues that came up as a result of the changes because sometimes Gemini would have the syntax slightly off or I forgot to include a file so Gemini guessed what that code would look like.

The entire process is really close to what Claude Code, Gemini CLI and Codex are doing these days, but I’m adding the manual check and copy-paste steps in myself rather than letting the AI handle everything.

Many times the thought crossed my mind that I could let the models run on their own, hook them up with a Scrum board and let them knock out tasks in the background, but I haven’t committed to setting up that system yet, and I don’t want to pay for it at this point.

Most likely the cost of a full coding agent would only be a few dollars but it would depend on the scope of the changes and how accurately it solves the problem.

Gemini CLI will probably be the first one I try since I think you get 1 million tokens for free, which is a substantial starting point.

Once I know a project will have more of a budget, I’d be much more willing to let something like Gemini have more autonomy to work in the background without me.

I’m sure I’ll hook it up and give it a shot soon.

But for now, it’s wild to reflect back on where Wave started and the state of AI.

I started Wave with VS Code, but very early on switched to trying Cursor, which was just taking off.

And back then, Claude Sonnet and GPT o1 struggled to get anything beyond pretty buttons and screens to work, but more than 10 AI models and versions later, Gemini and o3 are producing fully featured, scalable, enterprise apps.

And Grok 4 is looking push the bar a little higher.

Who knows, maybe this next phase of Wave will be handled entirely by autonomous coding agents, and I’ll just be there to manage.

Next stop, AGI.

Grok 4

Speaking of AI improvements, Grok 4 released earlier this week.

I haven’t had much time to test it out, but I had it validate my YouTube and Skool strategy which it came back with a very solid answer, so that’s a promising start.

I didn’t have it do much by way of code though, so I’ll be anxious to see how it does with programming.

Grok 3 was a very solid coder so I would expect Grok 4 to be better.

Since it just released this week, it will be some time before it finds it’s place on the LMArena Leaderboard.

One interesting option I noticed with Grok last night was the “Grok 4 Heavy” option.

Naturally I had to see if I was able to use that option but sadly I’d have to pay more.

Seeing “$3000” was a little shocking at first, but “per year” helped, and doing quick math on that:

$3000 / 12 = $250 per month

$250 per month is right on par with OpenAI and Google’s pro tier subscriptions.

Unfortunately I’ll have to hold off, I can’t exactly justify $750 per month on AI services right now, so I’ll settle for regular old Grok 4, but I’ll be curious to see how Grok 4 Heavy performs compared to Grok 4 and the other models these days.

I’m sure I’ll have some coding objective to test Grok 4 next week so I’ll let you know if it overtakes Gemini for me.

Automatic Shorts Part 3

With the release of Wave this week, I got behind on the Automatic Shorts series on YouTube but the plan is to record it right after I finish this newsletter.

I’ve already gone through the steps of finding clips from a video based on the transcript, pulling out the clips into separate videos, and reformatting them to a mobile format, so the last part will be recombining all those clips back into one single video.

And based on a comment on YouTube, and something I already mentioned in the part 2 video, I may throw in subtitles if it doesn’t turn out to be too much extra time.

I’ve enjoyed the live building aspect of these videos so I’m thinking it will be a semi-regular thing going forward, alternating with more prepared builds every week.

The goal is to get up to 2 videos per week, which I think will be manageable barring any unexpected virality with Wave.

But with Wave in a good spot, and this Automatic Shorts YouTube series wrapping up, it felt like the right time to introduce my next big focus…

A new Skool community.

Learn Automation and AI

Originally GPT convinced me to wait until I had 1000 subscribers on YouTube before starting a Skool community, but I recently decided to get started early.

Grok 4 and Perplexity both reassured me I’d be fine if I started a group now, and I’ve been excited about the idea lately, so today is another soft launch, as if Wave wasn’t enough.

I’m opening the “Learn Automation and AI” community on Skool:

Kind of a boring name, but hard to mistake the goal!

Plenty of time to change it later, but I like it for now.

Later today I’ll release part 3 of the Automatic Shorts YouTube series I’ve been recording and wrap it up by introducing the Skool community.

Already I’m liking having a single place for all the videos and workflows, rather than being disjointed between Gumroad and YouTube, plus now having a better place for people to hang out, post fun stuff, and ask questions.

And I was enjoying setting up the community way more than I thought I would.

Looking forward to adding more helpful videos and posts and talking more about AI and automation, and providing some value along the way, while hopefully growing a community of AI and automation enjoyers like myself.

The plan is this group will be free and I’ll introduce a paid group update later with a few more benefits than the free group if you really like what I’m doing, but I’m still planning to have all the workflows and videos be free.

The goal is to provide as much free value as possible.

And who knows, depending on how this goes, I may have a second group focused more on vibe coding and everything that went into building Wave, but this group for now will focused on automating with n8n.

Check it out if you get the chance, and let me know what you think!

And that’s it for this week! Wave finally launched, and a new Skool community ready to go! Lots to do.

Those are the links that stuck with me throughout the week and a glimpse into what I personally worked on.

If you want to start a newsletter like this on beehiiv and support me in the process, here’s my referral link: https://www.beehiiv.com/?via=jay-peters. Otherwise, let me know what you think at @jaypetersdotdev or email [email protected], I’d love to hear your feedback. Thanks for reading!