App development, and too many AI options to choose from.

DeepSeek, Perplexity, and OpenAI, and I’m sure others, have been very busy this week.

What’s happening:

• iOS App Testing

• Digital Assistant Setting

• Operating Browsers

• DeepSeek Market Results

• OpenAI’s Response

iOS App Testing

Going strong on app development this week with another first.

I’ve deployed an app to the Google Play Store, but iOS is new territory.

Creating an installable development version of an iOS app is nearly as much work as fully publishing it, which is a little frustrating.

I feel like I should be able to deploy a test app from my own computer to my own phone without requiring a certificate, multiple approval steps, and a $99 developer account with Apple.

Expo Go allows for emulation through their app, but it’s always better to test an install directly on the hardware.

After a few hours spread over a couple days, I accomplished a first though.

A TestFlight version of an app that could be installed on iOS.

Took some trial and error, but with AI assistance, specifically Claude via Cursor and a handful of Perplexity searches, along with Expos EAS, I had a build created and submitted in just 2 commands.

One to build the .ipa file:

And one to submit the build (after logging into a newly minted Apple Developer account complete with certificate):

A little more configuration in the App Store Connection portal and there was finally an email from TestFlight to install an app on my device.

So simple!

On the Android side, the steps are as follows:

• run build command

• wait for build to complete

• download build

• install build

Granted, the initial setup for iOS was on the painful side, it’s relatively easy to create and deploy new builds now, with a few extra steps than Android.

And now I can officially cross compile an app for iOS and Android.

Well worth a few hours of struggle with huge help from AI and Expo.

Nearly ready to actually publish!

Digital Assistant Setting

I’ve been using Android for a long time, but somehow I’ve never noticed the “Default digital assistant app” setting until Perplexity offered to change it.

Opening the Perplexity app this week, they introduced their new Perplexity Assistant feature, which could take over as your phone’s primary assistant.

I decided not to make the change at the time, but it’s worth considering.

I don’t really use any assistant features on my phone to begin with, besides occasionally using a voice command if I have a headset on, but if the assistant has enough access to the device’s settings and apps, it could be a powerful option.

Similar to OpenAI’s tasks, Perplexity assistant could do scheduled search results compiled from multiple sources, or read through app notifications, emails and messages to provide briefings or prioritize the day’s activity.

One other benefit this setting now provides is versatility.

Perplexity usually lets you select a few different AI models to use, including the new DeepSeek R1 model, so you could potentially experiment with a fully reasoning model running your phone without having to wait for a major upgrade or needing to buy a new device entirely.

I might preview R1 as my assistant this week and see what happens.

We’ll see if this setting I didn’t even know existed becomes a game-changer.

Operating Browsers

OpenAI’s Operator feature has been out for more than a week now but I haven’t given it much of another chance yet.

The vibe online seems to be that it’s a bit restricted and still needs some guidance unfortunately but it’s a step in the right direction.

As a result a few other competitive options are taking advantage.

One promising project is Browserbase, a headless browser built for the same purpose as Operator, AI web navigation, but without restriction.

APIs still reign supreme at this point, and plugging something like R1 into browserbase is another way to give an AI full reasoning power to navigate the web and accomplish tasks on it’s own a little more accurately since it can think about the task first.

Operator will probably be able to do the same soon as more testing results come in to decide what to restrict, as well as rolling in o1 and o3 models to help reason itself on the right track and away from commands to DDoS, but this may be the way of the future.

Browsers might no longer be necessary in the not too distant future where AI’s can compile random data sources as is and display the data however the user wants to see it, no browser restrictions required.

Haven’t given browserbase a shot yet, but I’m sure as I wade deeper into automation territory, I’ll undoubtedly find myself in a scenario that requires a browser navigation workaround.

Another tool to patiently wait in the toolbox while waiting for OpenAI to eventually handle everything for me.

DeepSeek Market Results

DeepSeek caused a bit of a panic in the markets this week.

Supposedly R1 only cost about $6 million to train which is considerably less than $500 billion, $100 billion or $100 million which are the numbers that have thrown around lately for training these large-scale models.

An interesting theory on why it was so cheap to create R1 is that DeepSeek used ChatGPT to build and train a considerable portion of the model, including massive API calls to OpenAI coming from China.

It will be interesting to see what else unfolds as more details come to light, whether DeepSeek truly revolutionized model creation or if it was heavily assisted by GPT, but in either case I’m not surprised.

AI will be improving AI, and things are only going to move faster.

Hopefully AI companies are doing just that, using their product to improve their own product.

FireShip had a quick sarcastic breakdown of the situation below if you need a full recap:

OpenAI’s Response

With everyone talking about DeepSeek R1 and questioning OpenAI’s progress, OpenAI quickly responded today.

Just an hour before writing this, OpenAI began rolling out access to their newer model o3, barely making their Shipmas claim of it being available later in January.

At this point, DeepSeek basically forced OpenAI’s hand to try to prove their dominance.

Since it’s so new, I haven’t seen any performance metrics comparisons yet but I’m sure there will be plenty of chat about it next week.

The AI race is not slowing down anytime soon.

With that, I’m off to give o3-mini and o3-mini-high a test run to see what they can do since I just confirmed I have access to them.

And that’s it for this week! iOS app development and too much AI to keep up with.

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!