Huge week in AI as all the major players compete in big announcements.

A more grounded week for me as I try to push Wave to the finish line.

As always, much to discuss:

• App Publishing

• Operating Costs

• Google’s Lead

• Microsoft’s Play

• AI Developer Team

App Publishing

Since Wave isn’t published yet, now is the perfect time to migrate the app to be published by Wave rather than Divs Design as it currently is in TestFlight.

It sounds like Apple will let you switch publishers after the app is live, but it can be a slow and painful process.

Might as well take the extra time to publish the live app under the proper publisher instead.

But that means Wave needs an Apple Developer account to be able to publish and apparently Apple prefers you have a business domain email rather than a plain old @gmail for your developer account.

My old approach to creating an online business foundation used to rely on creating a gmail account to get things started and in some ways that’s still the foundation.

Unfortunately I think too many people abuse process as well, which has resulted in Apple preferring a proper email with a custom domain to be the publisher.

I’m glad I checked that before signing up because that’s exactly what I was about to do, use the business admin @gmail account to sign up for Apple Developer account.

One more headache avoided, setting up a new business branded email address to use for apps instead.

With that, I believe I have all the accounts needed to recreate the publishing process under a new account.

I’ll have to setup new certificates and change all the values in the app to use the new accounts but the app itself should remain unchanged thankfully, only minor code and config tweaks.

Once I’ve got the publishing process up and running again, the Wave app will be ready to finally go live.

Assuming I don’t get completely distracted with all these shiny new AI announcements…

But next up the, cost of operations.

Operational Costs

It is a little scary how quickly the costs of operating an app can add up.

I’ve been doing my best to keep costs in mind and I’m sure there’s further optimizations that can be made, but usually those cost saving tactics come at the price of time and security.

Self-hosting is usually the lowest cost solution, but that’s such a huge headache to have to manage authentication and networking and all those other elements that it makes more sense in the long run to pay for a service that can relatively cheap provide a managed solution to most of those problems.

Right now, Wave already looks to cost over $100 per month to operate:

• $30 for a custom website

• $9 × 2 for custom email addresses

• $30+ for Supabase (with custom auth url)

• $24 for one full AWS App Runner service

Plus $99 per year to license an Apple Developer account.

As I mentioned last week, App Runner is a little pricier compared to Lambda’s free tier but Lambda should be good until we have 1000+ users bouncing around the app, creating consistent traffic on those services.

And of course, video can get really pricey really quickly if not managed properly, but luckily I believe I’ve created a decent runway to handle that for a while.

The solution to all of this is make more money than you spend.

We have a few ways to monetize the app but just need to get those ideas into the hands of users to validate they are valuable enough to pay for.

In the meantime, Wave may cost a decent amount of money to operate.

It’s amazing there are so many free apps out there because apps are not cheap to run.

Google’s Lead

Google’s “I/O” conference had a ton of AI announcements.

One thing I found funny was they seem to embrace the LMArena leaderboard as much as I (and many in the AI community) do, using multiple screenshots to highlight that Gemini Pro 2.5 has been at the top in multiple categories.

Their new model Gemini Flash 2.5, which is supposed to be the cheaper, faster model option compared to Gemini Pro, is number 2 on the ranking right now:

Google’s AlphaEvolve balances between those 2 models to more efficiently find solutions to algorithmic improvements, and possible improve those models directly.

Beyond just showcasing their leaderboard standings, though, they announced a number of upgrades and new products including:

• Imagen 4 - upgraded image model

• Veo 3 - more realistic video generation model that can also generate sound

• Flow - an AI video editor

• Lyria 2 - an updated music generation model

• A “Deep Think” mode for Gemini Pro which I didn’t realize it didn’t have yet

• Jules - an async developer mode that can work on code independently, similar to OpenAI’s new Codex

• A separate “AI Mode” for Google search now so they don’t cram an AI result at the top of the search result anymore

• Project Mariner for research and browsing the internet automatically

• Project Astra for more a realtime assistant

• Glasses integrations similar to Meta’s glasses and other Android XR projects

• Other things I missed

Of all that, this video is probably my favorite:

They closed all that out with a nice pricing update, the standard $20 per month subscription to get minimal access to some of those features, and a hefty $250 per month to access all of those new models and options.

Not sure I can swing multiple $200+ per month AI subscriptions at this point between OpenAI and Google but it is tempting to see which one proves more useful.

I’m sure I’ll be starting a free trial soon to see if I can’t live without all these new upgrades.

And here’s a full cut of all the Google announcements on Wes Roth’s channel:

Microsoft’s Big Play

To compete with Google, Microsoft went the other direction and completely open-sourced their Copilot app.

Rather than trying to charge $200 or $250 per month like OpenAI and Google, Microsoft is banking on the platform-approach long play.

Give most of it away for free and let your users improve the process for you, but of course also have easy integrations into Windows and other tools so users would be more incentivized to adopt the Microsoft ecosystem.

At the same time, the bigger part of Copilot’s AI lives on Microsoft servers which you’ll have to pay for compute, but the rest of the Copilot code is now free to use and modify as needed.

Get the developers to adopt your code rather than Google’s approach of creating an all around customer approach.

OpenAI’s approach remains somewhere in between, but maybe not for long as they start to roll out their own devices after the acquisition of Jony Ives io… but that’s for another day.

Time will tell if Microsoft made the right move, but it seems like a smart play.

AI Developer Team

OpenAI announced their own async coding agent last week, OpenAI Codex, which is similar to Google’s Jules, Microsoft’s Copilot Agent, and Anthropic’s Claude Code.

It basically tries to act as your hired code contractor that can connect directly to your code base, be given a task and operate in the background creating updates ready for you to approve when it’s done.

The idea is basically here, a full team of AIs doing all the development, constantly churning in the background to fix bugs, write tests, make minor improvements and develop small new features with little guidance.

It’s weirdly enough, not too different from my day-to-day at Salesforce, working with real people.

I was usually in meetings, but I would be answering questions through Slack and checking on work item progress and scanning PRs throughout the day to make sure the things we said we would do got done, and done correctly with proper tests and reviews, and made sense with the current goals and scope.

I’d jump on a call with someone hear and there for clarity but most of the time I’d answer questions in chat and they’d be good to go.

AI can basically operate the same way in this capacity, a full-time remote employee working on code tasks you assign to it and checking in when it gets stuck or needs more information about the particular work item.

The biggest unlock with these newer asynchronous code processing tasks is that these companies are now providing the AIs with their own computers to do the work rather than you having to allow them to do the work on your computer.

This has slipped under the radar for me as I’ve been focused on getting things done, but I’m going to have to come back and try out some of these tools.

It may be time to hire out AI systems to see what they can do on their own.

Bonus: More AI!

A couple more AI things I realized I missed as I wrap up.

Grok was upgraded with live internet searching to compete with all the other major models.

Not surprisingly, Grok has been very strong at searching X/Twitter specifically for recaps about what experts are doing on the app (and cutting through most of the noise on that platform), so it will be interesting to see if it proves just as powerful with the rest of the internet available for search, which I think it was already starting to do…

But the big last minute addition here is Anthropic just announced Claude Opus 4 and Sonnet 4.

Claude has been a go-to model for me in the past, but as other models have outpaced it, I had left it out of the rotation.

I’ll be anxious to see how these new versions of Claude turn out.

Sonnet 3.7 has remained in the AI leaderboard top 10 for a few categories like WebDev and Copilot but we’ll see if this new version catapults it back to the lead in other rankings.

Lots of AI testing to do next week.

And that’s it for this week! Still preparing for an app release and trying to not get completely distracted with all the AI announcements. Much 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!