An Android slow down.

Lots of AI updates from all the big players.

And plenty more to do as always.

Topics for this week:

• Play Store Stalling

• Brand Stats

• Nano Banana

• VibeVoice

• More Voices and Codexing

Play Store Stalling

I wasn’t expecting this.

The Google Play Store is taking a surprisingly long time to approve the app.

I did get bitten by vibe coding again, though.

Classic AI move using out of date training to recommend old APIs.

I submitted the Android build of Wave on Tuesday, and they came back with one issue that I was using an API version that was too old, so it was easy enough to update, recompile and re-submit.

But that was on Wednesday and as of this Friday morning it’s still under review.

So far, Apple has been faster at reviewing than Google, which I would not have guessed.

Historically Google has allowed for some pretty nasty situations like fake celebrity apps and conspicuously huge “flashlight” apps that turned out to be viruses or backdoor admin access so maybe they have tightened up their review process in the last few years.

Meaning sadly, I don’t have an Android link for Wave yet.

But I can add you to the insider testing if you’re interested.

Hopefully next week, Google will be happy with the app submission and Wave will finally be live on both mobile platforms!

Brand Stats

Slow and steady progress.

I’ve been tracking Skool Activities in a daily spreadsheet but I’ve since expanded that to tracking Skool Members and YouTube Subscriptions in their own sheets this week too.

It has become the starting point for every livestream now.

It’s fun to see the numbers everyday, plus it gives an idea of how things are going, eventually leading to better decision making because I see what’s up, down or sideways.

And like anything, it does require balance.

Obsessing over the numbers usually hurts the numbers, but operating without any numbers leads to guessing on what direction to go next.

Easy to say this now because conveniently my numbers have been pretty good!

Amazing what being consistent can do.

Just this week, up 30 Skool Members, 24 YouTube Subscribers, and finally got my Skool Activity average over 20, which is the first milestone.

Next stop, 200 average Skool activities per day.

Need to get all the other platforms in the mix eventually - TikTok, Instagram, LinkedIn, X, but slow and steady for now.

4+ live streams and 1 full YouTube video a week is enough to manage as is.

Speaking of YouTube videos…

Nano Banana

The infamous “nano banana” image model dropped this week from Google.

Gemini 2.5 Flash Image is now available to preview on aistudio.google.com.

And to celebrate, I made a video on how to spin up a simple image editing app with n8n:

The so-called “Photoshop killer”, Gemini 2.5 Flash Image can quickly and accurately edit images based on a prompt like “change the sweatshirt to yellow”.

Another system I’d like to try out is one based on it’s ability to accurately combine images.

This example from Google is really impressive:

Easily combines 3 different images into a realistic result, including a reflection of the person on the wall.

I may build another n8n workflow to do something like this next week.

Let me know if you end up trying out “nano banana”.

VibeVoice

Another pretty crazy release this week was Microsoft’s open source VibeVoice.

It can generate realistic sounding voices similar to NotebookLM, but it takes it a few steps further by producing 90 minute audio segments, and it can create a conversation between 4 different speakers within the same audio file.

I’m mostly surprised this is near NotebookLM quality available to use for free since it’s open source.

I haven’t had a chance to try it out myself but from the clips I’ve listened to from their website, it sounds really solid.

This could become the go-to replacement for ElevenLabs by enabling high-quality audio generation for free, the same way Whisper turns speech to text for free.

I haven’t seen too much hype for this online yet, though.

Nano banana may have stolen the thunder, but I’m sure we’ll be seeing more Vibe Voices soon.

More Voices and Codexing

On the paid side of things, OpenAI released 2 major features this week, both developer focused.

They did a quick livestream to show off their new Realtime API for speech-to-speech capabilities.

The voice itself is a little dramatic, but also kinda mindblowing how it can now match emotion, fluidly switch between languages, and be as responsive as it is.

I spent at least 15 attempts at one point trying to get Eleven Labs to generate a single voice line with the proper inflection and still never figured out how to get it to sound like I wanted, but this seems like it can get way closer in one shot.

There’s a slight delay in it’s response time but I’m sure it will only get better in the future:

On top of that, OpenAI expanded it’s Codex functionality this week.

They released Codex CLI a little while ago to match Claude Code and Gemini CLI, and on Wednesday they added an IDE extension, cloud and local swapping, code reviews and an updated CLI.

I keep meaning to try more of these CLI based systems, especially for background processes, but I’ve been hesitant to connect a credit card to Gemini or Claude only to have them spin themselves into a nice little logic loop where they burn $1 in tokens in a few minutes doing things I don’t even want them to do.

But since I do have a GPT subscription, this might finally push me to try.

I heard a great use case for it the other day too.

Often times when coding, I’ll be in the middle of fixing or building one feature, when I spot some other small issue.

I’m not the best but I try to keep my code commits focused on one general goal, and not mix in fixes that are unrelated to the task.

But sometimes I cave and fix the issue because I don’t want to forget about it, or don’t want to go add an item to the backlog to come back and fix it later and then have it deprioritized for other tasks.

With something like Codex already connected and ready to go, I could quickly tell it to go work on the issue in the cloud on its own and create a Pull Request for me while I keep focusing on the task at hand.

Once I’ve finished my part, I can check what it did for the fix.

That is 10x engineer stuff right there.

Hopefully it ends up working like that anyway.

I’ll let you know if I finally end up hiring those AI contractors I’ve been talking about or if I stick with my de facto vibe coding methods.

And that’s it for this week. Big updates from all the big AI players. Speaking of, “sonic” was revealed as grok-code-fast-1 yesterday too.

Let me know what else I missed!

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.

I also have a Free Skool community if you want to join for n8n workflows and more AI talk: https://www.skool.com/learn-automation-ai/about?ref=1c7ba6137dfc45878406f6f6fcf2c316

I’d love to hear your feedback at @jaypetersdotdev or email [email protected].

Thanks for reading!