It’s been a long week.

Lots of roadblocks.

Luckily some work arounds too.

Topics for this week:

• Nearly Released

• Divs Denied

• App Store SMS

• Misjudged

• Good News

Nearly Released

Not surprisingly, I don’t have a link to the app yet.

A few last minute changes, maybe an over-reliance on AI, and a touch of gold-plating led to finally getting a distributable version being built, but not an App Store version yet.

The features were talked about before so they weren’t too much of a surprise.

There was a stretch where I was letting Claude handle too big of a scope of changes which ended up taking a while, so could have saved some time there.

And if I didn’t learn my lesson from Claude taking too long, I did give it a few attempts to generate a nice-to-have feature but I went in planning to shelve the idea after 4 attempts.

Luckily it was able to deliver better that expected on that.

After all that, there were still a few other hiccups trying to set up an email system and then creating a build for iOS but with all that, I think we’re nearing a close to this MVP.

Next week.

Next week I’ll still be talking about whether this app made it to the App Store or not.

But hopefully I’ll have a new version installed on my phone finally.

And also find some time to work on automations again…

Divs Denied

I’m a little worried I set something up weird with Divs Design.

Supabase barely “supports” email for their authentication services.

They’ll send 2 emails an hour.

Clearly they want you to use another email provider so they recommended SendGrid.

I made it through the brief account signup process on SendGrid when I was suddenly kicked out.

I had entered all the typical account credentials, and made it to the step where they would validate that I own my domain, when the page reloaded and I couldn’t log in.

Maybe 5 minutes later, I received an email from SendGrid Support that they had determined I would not be able to proceed with activating my account.

I attempted to ask for more information and offered to provide any extra details to prove I’m a human that owns my own company and domain, but they would only send me templated responses referring back to Policies saying that was all they could do, and each time they automatically mark the case as Solved, so clearly they aren’t going to try to help me at all.

One more try…

Apparently I violated something in their extensive “Acceptable Use Policy” or “Email Policy” but since it’s security related, they could not tell me what.

The only clue I have is from the first email they sent, referring to my accountId as “unified_acct_US…”

I forgot that I signed up for Twilio a little while ago to test out sending text messages from an app, so technically this would be my second Twilio account registered from the same device because I forgot Twilio owned SendGrid.

But even if that is the case, I’d rather have a business account be separate from personal, and I’m not sure how having identified a duplicate account would be a policy violation, especially when my other account still works.

Merge, connect or “unify” the duplicate accounts, and cool, now you know this one person has 2 different accounts.

If they try something malicious on either one, ban them both.

Not sure sure it warrants rejecting the activation process all together, unless I’m missing something.

Having been on the inside of overly complicated enterprise platforms like this, I would bet money there’s at least 2 developers somewhere in the organization that could identify this issue as a false flag, or know that this situation was swept under blanket logic because it wasn’t valuable enough to figure out what to do in this situation when there are so many other shiny new features to sell half-baked because the launch date was determined long before anyone considered the technical feasibility.

Best case Twilio has had some terrible issues with duplicate account exploitations so they’ve locked down account activation with their “stringent security measures”, which I would understand.

Less phishing emails and texts is a good thing.

But the worst and more likely case - this is just a bug.

It’s been a long week, so I’m more frustrated with this than I normally would be, but at the same time I’m a little shocked at the brute force solution.

I’ll give them the benefit of the doubt, though, I probably did something wrong.

I’m hoping this is merely a case of overly restrictive account policies and not an accidental misconfiguration with Divs Design that’s going to get my email flagged with issues on other platforms in the future.

Luckily MailGun worked first try so I guess I’ll be using them instead... for now.

App Store SMS

I’ve talked about how great Expo Application Service (EAS) is a few times now.

I’ve used their build submit commands 6 times so far to create new iOS app builds, as recently as February 8th, with no issues.

But for some reason I ran into an issue on Thursday.

Since November, Apple and EAS have had issues with using SMS (text messages) to authenticate accounts with App Store Connect.

Looking online, some are saying Apple is moving away from using SMS as a way to authenticate with the App Store Connect account, but will require direct device authentication, where Apple sends the code directly to authorized devices, not through SMS.

So I started getting an error that the build process couldn’t complete because I couldn’t authorize my Apple account through SMS.

This situation seems to be somewhere in between, which is frustrating because it’s unclear if this is an EAS problem or an Apple problem and neither have provided answers.

EAS doesn’t offer an alternative in this current command structure, at least I haven’t been able to get it to set a different way of authenticating other than SMS, so the build command fails with the same error for every attempt I’ve made, even after following EAS’s instructions to delete old files and install the latest versions.

Apple hasn’t said it doesn’t support SMS authentication, but there are a number of people running into the issue and reporting in the EAS repo.

Ultimately I ended up creating a new API key in App Store Connect, and updated EAS to use that new key instead.

Then, when EAS would ask if I wanted to sign into my Apple account, I would normally have said yes, letting them connect to my accounts pull the credentials needed to create and submit the new app build.

This time I said no, and it seemed work to fine without trying to log into my Apple account.

Easy enough solution, but weird timing and another hour of searching to decide what went wrong with something that had worked 6 times flawless before.

I was finally rewarded with waiting in a 310 minute free tier build queue…

In the past that number had been an over-estimation and would quickly drop after after minutes, but not this time.

With 200 minutes remaining, I paid $.03 to jump to front of the queue and the build completed in the 10 minutes, and I sent the build off to Apple.

As of today I’m still waiting for them to approve the build for TestFlight but it hasn’t been 24 hours yet… 🤞

Misjudged

I mentioned it earlier that I keep wanting to trust AI fully but it’s not quite fully trustworthy.

I ended up switching back to Claude over Grok this week because my Cursor usage reset for the month and it’s way more convenient to use the built in AI chat in Cursor rather than constantly pull files out and paste them into Grok.

And in most cases Claude Sonnet 3.7 and Claude Sonnet 3.7 Thinking have been really solid.

Multiple times I asked for something and it delivered above and beyond.

But other times it was not as great, struggling to find the issue or understand what I was asking.

I did find that it seems to hallucinate pretty quickly, and Cursor noticed it too.

If the current chat becomes too long, Claude will start making wild changes or get distracted with unrelated ideas pretty quickly, so Cursor prompts you to start a new chat for better results.

Before I figured this out, at one point, after more than an hour of multiple cycles, round and round, I was trying to have Claude help me fix search filters, showing Claude the current error, letting it try to fix the issue on it’s own, only for it to get the same error or create brand new ones.

I tried to thoroughly explaining the issue, completely outlining the expected result and providing sample data to directly show what I was wanting to happen, only to have it implement an obviously inefficient solution that still didn’t work.

With that, I decided to turn the problem over to GPT 4.5.

I took the search query I was working on, pasted it into the chat box, and GPT spit out a way better solution that worked first try.

After removing all Claude code attempts, I verified a filterable search that I could have had a few hours earlier if I had just let GPT 4.5 help me come up with the initial query, but after 6+ hours of working on a project, it’s hard to fight the urge to let Claude take the lead.

During this, I had been using GPT 4.5 for one off, best practice questions and I had been really impressed with it’s answers.

It doesn’t fully reason, but seems to be somewhere in between.

It’s slow to respond so there’s probably a small element of reasoning in there, but it’s also been so heavily trained, it knows how to generate well formed answers.

I think 4.5 is still limited to more direct questions, it doesn’t seem to do as well with multi-step or reasoning, but it’s great at providing smart answers, not just doing what it’s told.

Last week, everyone was talking about how 4.5 was a failure, but I think it’s actually one of the best for what it is, and unsurprisingly it’s fighting for #1 spot in the Chatbot Arena Leaderboard, despite it not being able to reason as well as other models.

Good News

Somewhere in the middle of fighting with SendGrid, EAS and the App Store, I did get some good news.

My OpenAI account was granted API access to the full o1 and o3-mini models which previously had only been available through chat for me.

Despite fighting with Claude all week, I was still excited at the prospect of testing out more AI models.

Maybe a ridiculously expensive to operate automation flow that connects to all these different AI models will be in the near future.

Can’t help but think of all the untapped potential.

In the meantime, I’ve got an app to ship. And maybe a nap.

And that’s it for this week! A few roadblocks this week but what’s life without a little challenge.

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!