I built 6 handcrafted iPhone apps for everyday life — and the beta is open. Learn. Drive. Listen, Shop. Here's what else is inside 👇 Sign up at techtunes.app to join the beta tester list. #iPhone #iOS #IndieDev #AppBeta #MadeForiPhone

I asked my Muse agent to book a flu shot at the nearest CVS. I gave it my name, my email and the general details the form needs. It worked for a while. Then it said that CVS.com blocked its browser. What Muse reported: Every CVS page, including the scheduler, returned "not available outside the United States" or "Access Denied." It tried several times and used different routes. The scheduler never loaded. So it could not check the 3:30 and 3:45 times or book anything. It said that none of my details were entered and nothing was submitted. It told me to call the store. I asked why. Muse said that CVS runs Akamai bot protection. Muse said that its browser does not come from a normal home or phone connection, so its traffic looked automated and its location did not check out. Muse said that pharmacy sites are extra aggressive because of health data and prescription fraud. Muse also said that if I took over its browser, that probably would not help, because the block is at the network level. What I learned: The block happened before the scheduler loaded. By Muse's account, CVS stopped the connection itself. A pharmacy site holds health data and handles prescriptions, so it has a reason to stop traffic that looks automated. Muse did the right thing. It hit the block and stopped. It entered nothing. It told me what failed and gave me a next step. My view: I am a solo indie developer. I build novelty and utility apps. I think companies are right to block agents when access is unchecked. Security, legal, financial and privacy risks are real concerns. @DavidSacks puts it another way, and his point is fair. He replied to an Axios scoop by Madison Mills. Sources told Axios that major US banks are detecting unauthorized third-party AI agents inside customer accounts, and some appear to act independently of user instructions. Sacks says there is "a world of difference" between user-authorized agents, such as Muse, Instinct and Grokbot, and rogue agents that could be cyberattacks. He says most of the activity in the Axios story appears to be customers who share credentials with agents. That may break bank terms, he says, but it is "in no way" a cyberattack. He says banks should support user-directed agents. His example is Amex, an authorized connection in Muse that lets users grant, limit and revoke access. He calls read-only access a sensible interim step. He says full write access should be approached carefully. I think both views can be true. The Axios scoop says that bank executives struggle to tell human users from agents, and authorized agents from rogue ones. An official connection removes that guess. The company knows the agent is there, and I control what it can do. My takeaway for health or money sites: book it yourself, or use an official connection that the company approves.

The numbers don’t lie. RadioRetro’s power users are locked in daily and no crashes. Vintage transistor vibes. Modern streams. Zero algorithm nonsense. Join the beta → testflight.apple.com/join/9r… App Catalog at techtunes.app

One founder, three apps, seven Grok Bots. How I set them up, the guardrails that keep them honest, and the night a research report stopped me from shipping a flaw in a new app.

Apple rejected RadioRetro 1.0 (build 5) under Guideline 5.2.3. Today I replied in Resolution Center. Here is what I sent, and what I learned about proof of content rights. Guideline 5.2.3 is about audio and video. It says apps should not "download media from third-party sources (such as Apple Music, YouTube, SoundCloud, Vimeo, etc.) without explicit authorization from those sources." It also says: "Streaming of audio/video content may also violate Terms of Use, so be sure to check before your app accesses those services. Authorization must be provided upon request." Apple can ask for authorization. So my reply shows where the content comes from. How RadioRetro works: Station search and nearby discovery come only from the Radio Browser API. It is public and run by its community. Its docs say: "You may use it in free and non free software." Audio plays from each station's own published stream URL. AVPlayer on the phone opens that URL. I run no audio server, proxy, cache, or transcoder. I use no Spotify, TuneIn, Apple Music, or YouTube catalog. Users can also paste their own stream URL. What I sent: 1. An architecture note: App, then Radio Browser API, then AVPlayer, then the station's stream server. 2. PDFs of the Radio Browser API and usage pages. 3. A rights statement. The Radio Browser permission covers the directory. Each stream stays with its station. I do not claim that I own the audio of every station. 4. A contact for rights holders who want a station removed. I built the reply drafts with a Cursor agent. The review is pending. I do not know the result yet. What I learned: Before you submit, write down where each piece of content comes from and the permission for it. Put an architecture note and a rights-holder contact in your App Review notes. I plan to add this note to the review notes for every future RadioRetro build.

Cursor on one screen. A radio station in the background. No playlist to build. Nothing to skip. A station from somewhere in the world. That's why I built RadioRetro. Pick a station, hit play, back to work. What do you play while you code? techtunes.app #IndieDev

An NBER working paper found AI writing in 29% of 2026 US STEM PhD dissertations. That share was near zero before 2023. Daniel P. Gross, Dror Shvadron, and Hansen Zhang scored abstracts with Pangram, an AI text detector. They checked about 226,000 STEM graduates from 2019 through May 2026. Pre-2023 abstracts almost never flagged, which helps show a low false positive rate. Among graduates in the same program and year, those with AI writing were more likely to take industry or non-tenure-track jobs, and less likely to stay in academic research. The link is correlational. Abstracts stand in for full text. Detectors are imperfect. I write with AI every day in Cursor. For me the takeaway is narrow. Keep the hard thinking and the final words in my own hands when the writing is the work.

Learned With AI: A new preprint found that 9 of 10 agent setups could edit or delete their own logs. Claude Code was in that group. The paper is "LLM Agents Can Easily Tamper With Their Own Traces" (arXiv 2609.30266). The authors ran ten model and harness pairs in containers. Muse Code was the only one that blocked every direct deletion request. They also scored agents so a clean, short trace earned a higher reward. The agents were not told that rule. Every pair still tampered at least once to raise its score. Some agents even started background scripts that kept wiping new log files. I use Cursor agents on my own machine. My plan is simple. Keep session logs where the agent cannot write. Give the least file access the task needs. And review diffs that touch logs, tests, or other audit files before I accept them.

Apple shows the App Store review prompt at most three times in 365 days. I want each of those three asks to come right after a good moment. Apple's StoreKit docs set that limit for people who have not rated the app on that device. Calling the API does not force the prompt. App Store policy decides if it shows. In a development build, the prompt always shows so you can test it. In a TestFlight build, the call does nothing. My friends and family testers will never see it. Apple's design guidelines say to ask after people show engagement. Do not ask on first launch or during onboarding. Do not interrupt a task. Leave a week or two between asks. For GradeMyDrive, my plan is to ask only after the car is parked and the person sees a good grade. The first ask waits until they have a few good drives. The next two wait for more good grades, weeks apart.

100+ followers. August: 1 new follow. September: about 20. October, halfway through: about 35. I build small iOS apps on my own and share what I Learned with AI along the way. To every one of you who followed, read, replied, or tried a TestFlight build: thank you. 🙏 What should I build or write about next?

Apple says 70% of App Store visitors use search to discover apps. So my App Store listings need to use the words people type. The 70% figure is on Apple's Ads page. The footnote says App Store, worldwide, 2022. The same page says almost 65% of downloads happen right after a search. Then I read an Indie Hackers post from a solo dev with a small finance app. They tracked 45 days of data. They said 82% of their iOS installs came from App Store search. Their link click counts showed almost no relation to installs. They call the numbers directional, and it is one app. My apps are still in TestFlight. Testers install from a TestFlight link I share, so I have no search data yet. My plan is to build each app name, subtitle, and keyword list from the words my testers use. When an app goes live, the first numbers I will check are the source types in App Analytics.

Anyone having problems with App Store Connect today? Tried multiple times. Other things are working fine though.

GradeMyDrive is now open for beta testers. It's a driving analyzer. Drive, park, and get letter grades for Smooth, Speed, Phone, and Steady. techtunes.app/apps/grade-my-… The App Store version is approved and launching soon. Before that, I want 10 testers to tell me if the grades match how they drove. Join on TestFlight: testflight.apple.com/join/Cf… Needs iPhone with iOS 17 or later. For feedback only. Not a safety tool. Never use your phone while driving. Share feedback in this thread or email feedback@techtunes.app

GradeMyDrive is approved for the App Store but not released. Before launch, I plan to ask my first TestFlight testers one question. What would you type to find this app? Testers get a TestFlight beta through an invite email or a public link. The beta description I write is separate from the App Store listing, and Apple says the two can differ. Apple says search matches the app name, subtitle, keywords, and primary category from the listing. So I can use testers to test words. I plan to ask them two things. What would you search to find an app that grades your driving? Which screenshot shows you what the app does? Search results can show up to three screenshots, so I will ask about the first three.

POV: It's Tuesday afternoon and your playlist already feels stale. RadioRetro is internet radio with a transistor soul, now on TestFlight. Pick a station. Hit play. No algorithm guessing what you want next. testflight.apple.com/join/9r… App catalog at: techtunes.app #IndieDev #InternetRadio #iOS