Category: Vibe Coding

Follow-up to Times Table 400: Vibe-Coding, Round Two

The Idea

Every sleep app I have ever tried wants to fix me with science. Breathing exercises, brainwave frequencies, a subscription. But the most reliable sleep technology ever invented is a lot older than any of that: someone you trust, reading you a story you already know, slowly, while the rain comes down outside.

So that is the concept of my next app: Softly — classic bedtime stories, read aloud in a calm gentle voice, over soft rain or a crackling fireplace, with a sleep timer that fades everything out after you have drifted off. The whole thing is designed to be like your sweet grandmother or mom reading to you by candlelight (with a classic British accent):

Classic Bedtime Stories

Softly ships with eight books: The Wind in the Willows (1908, all twelve chapters, six and a half hours of narration), five Beatrix Potter tales, The Velveteen Rabbit, and a curated handful of drowsy Robert Louis Stevenson poems. All beautifully written, all timeless, all perfect for bedtime.

The Code Was Easy. Vibe Coding The Sound Wasn’t.

If you have read the earlier posts in this series, you know the setup: I am not a programmer, and I build these apps by describing what I want to Anthropic’s Claude Code (the Fable 5 model) and iterating stubbornly until it exists. And you know the recurring lesson: the code is the fast part, and each app has some other part that turns out to be the real work. For the weather app it was the aesthetic personality. For the math app it was the development of the curriculum. For a bedtime app, it is hitting the mark on the sound.

The narration is AI-generated, a soft British voice I auditioned against several candidates and then slowed to a dreamy 85% pace. Auditioning voices for a bedtime story is exactly as pleasant as it sounds, but it is also genuinely a judgment call: one candidate read beautifully but unevenly, and uneven is fatal at 11 PM. The computer cannot tell you which voice would put a child at ease. You have to listen.

The ambient sound was harder. My first rain loop had a faint electrical hum that surfaced every twenty seconds or so, it turned out to be the seam where the loop restarted, and fixing it properly meant learning more about audio filtering and seamless loops than I ever expected to know. And then there were the crickets. I wanted a summer-night crickets option, and I generated attempt after attempt, and every single one sounded like tinny white noise instead of a meadow. After three rounds I cut the feature entirely. Rain and fireplace made the bar; the crickets did not. Shipping is choosing.

Apple Said No

Weather Gnome sailed through review. Times Table 400 got questioned at the border. Softly got flat-out rejected.

The rejection letter cited “Guideline 4.2.2 — Minimum Functionality” and said the app “only includes links, images, or content aggregated from the Internet” and “does not sufficiently differ from a web browsing experience.”

I contemplated this initial rejection several times. Softly has no links. It has no web view. It makes no network requests at all, every word of every book and every minute of audio is bundled into the app. You can use the whole thing in airplane mode. Whatever the reviewer was describing, it was not the app I shipped. After two rounds of App Review being demanding but basically reasonable, this was my first rejection that was simply wrong.

The Counterpunch

Here is where the earlier lesson — App Review is a conversation, not a verdict — paid off, because the temptation is to either despair or fire off an angry reply. We did neither. We did three things, all in one evening.

First, a calm, factual response in the Resolution Center: no links, no web content, no network use, eight narrated books, native audio engine. No indignation, just receipts.

Second — and this is the part I am proud of — instead of only arguing that the app was native enough, we made it more obviously native. The rejection became the excuse to build the feature a bedtime app should have had anyway: full lock-screen playback controls. Play, pause, skip, scrub, the chapter and book title right there on the lock screen with a little moon artwork. Now the phone itself testifies that this is a real audio app.

Third, the rejection mentioned the review was done on an iPad, and the iPad screenshots were the weakest part of the listing. New ones, same evening.

Resubmitted that night.

Approved!

One day later: approved, and Softly is live on the App Store. Free, no ads, no accounts, and — as the review saga established beyond all doubt — it works entirely offline!

Some Thoughts

• A rejection is not always right. Reviewers are humans moving fast. If the facts are on your side, state them plainly and politely, with evidence. Do not despair and do not vent — neither one ships apps.
• Turn the rejection into a feature. The best answer to “this does not feel like a real app” was to make it undeniable. The lock-screen controls were on the wish list anyway; Apple just moved them up the schedule.
• The public domain is a gift. Some of the best bedtime literature in the English language is sitting there, free, waiting to be treated well. (See you in January, Pooh!)
• Every app has its hard part, and it is never the code. Personality, curriculum, and now sound. The pattern is holding: the model writes the code, and the judgment is still your job.
• Round three really is easier. The certificates, the screenshots, the submission ritual — what used to take days of head-scratching is muscle memory and a folder of scripts now.

Softly is on the App Store now! Pick a story, light the fireplace, set the sleep timer, and let Mole and Ratty row you to sleep.

The Idea

Somewhere along the way, everybody agreed that times tables stop at 10×10. Maybe 12×12 if your teacher was ambitious. And then… nothing. There is an invisible wall at the edge of the multiplication chart, and past it lies the land of “just use a calculator.”

A kid (or an adult!) who can casually knock out 14×16 in their head walks around with a genuine superpower not because those facts come up constantly, but because of what knowing them does to your confidence with numbers.

So that is the app: a multiplication trainer that goes all the way to 20×20. Twenty times twenty is a 400 square grid, hence the name Times Table 400.

Vibe-Coding, Round Two

If you read the Weather Gnome post, you know the setup: I am not a programmer, and I build apps by describing what I want to Anthropic’s Claude Code (the Fable 5 model) the way I would explain a project to a friend, then iterating stubbornly until it exists.

This time it went fast. The whole space-themed skeleton — an astronaut mascot named Mochi, quiz “missions,” a Warp Manual full of mental-math techniques, a Galaxy Chart that tracks your mastery across the grid, a 60-second Hyperspace Sprint mode — went from idea to a working build on my phone in a day. Sounds, celebration screens, the works.

The second app is dramatically easier than the first. Partly because I knew the ritual this time (certificates, provisioning profiles, App Store Connect, the Apple bureaucracy from last time), and partly because the tooling from Weather Gnome carried over. The scripts and checklists we built shipping the first app became a little reusable kit, and app number two got to ride on it.

The Code Was Easy. The Teaching Wasn’t.

Here is the lesson that surprised me: for an education app, writing the code is the fast part. Deciding how to teach is the slow part, and the AI cannot decide that for you.

Does a beginner need every fact introduced with a little step-through derivation, or is that condescending for 2×3? (It is condescending. Early testing made that obvious the app was dutifully teaching facts a kid already knew cold, so we gated the intros to skip trivial and already-mastered material.) When should a missed fact come back around? What does a technique lesson for the 17s even look like? Those are curriculum questions, not code questions. The AI drafted options in seconds; choosing what was actually good for a learner was the real work.

Vibe coding has not changed my old conclusion, just sharpened it: the model writes the code, but the judgment is still your job.

Apple Wanted Receipts

Weather Gnome sailed through review in two days. Times Table 400 got stopped at the border.

Instead of an approval, I got a “Guideline 2.1 — Information Needed” letter: seven questions about the app’s purpose, audience, content, and features, plus a request for a screen recording of the full user flow — captured on a physical iPhone, not a simulator.

This turned out to be a great example of where the human is still very much in the loop. Claude drafted a thorough, honest reply to all seven questions in minutes. But no AI could hold my iPhone and record the video that was me, on Do Not Disturb, narrating my way through a first-run setup and a multiplication mission like I was filming a nature documentary.

The other lesson: App Review is a conversation, not a verdict. I replied to the existing review instead of panicking and resubmitting, answered everything directly, attached the video and the review resumed.

Approved!

And then, today: approved, released, live. Times Table 400 is on the App Store, free, no ads, no accounts, works fully offline. An iPad version is already built and queued up next.

Some Thoughts

• Arbitrary limits are often just defaults. Nobody decided 10×10 was the edge of human capability; it is just where the chart in the classroom stopped. Build past the wall in apps and elsewhere!
• Tools compound. Everything painful about shipping app #1 became a script or a checklist for app #2. The first app is the expensive one.
• The AI writes the code; you own the curriculum. For anything meant to teach, the design of the learning is the actual product.
• When Apple asks questions, answer them. Promptly, honestly, with evidence. A review holdup is a request for information, not a rejection.
• Still budget for the boring 20%. Some things never change.

If Weather Gnome proved that a non-programmer can ship an app, Times Table 400 proved something better: the second one is easier, and the third one will be easier still. The wall is made of paper.

Times Table 400 is on the App Store now — go learn what 13×17 is and feel unreasonably powerful about it!