Apple allows volume subscriptions · How the iPhone Duo scales apps

The Mobile Takeaway
For people who ship apps
Friday, October 23, 2026

Apple just enabled volume subscription sales for businesses and schools through a new default toggle in App Store Connect.

Reviews of the new iPhone Duo show unoptimized apps look like legacy software with large black bars. MKBHD expects developers to need about six months to adapt.

Platform

• Businesses and schools can now buy your subscriptions in bulk Apple Developer News
Organizations can buy auto-renewable subscriptions in volume via Apple Business or School Manager. Multiseat purchasing is on by default in App Store Connect.
MKBHD artwork▶ YouTube · 29 min
MKBHD
Key takeaway: Apple’s $2,000 iPhone Duo shows unoptimized apps as small centered windows or with large black bars, and MKBHD projects about six months for developers to adapt.
• Most major apps, including Spotify and Google Maps, aren’t optimized yet and appear as small iPhone-sized windows on the 7.6-inch inner display.
• iOS 27.1 moves top and bottom controls to the right side of the outer screen for thumb access, leaving unoptimized apps with empty space where buttons should be.
• Instagram and Carrot Weather are among the first to update their layouts for the Duo’s aspect ratio and side-aligned controls.
• Users can lock two side-by-side apps into a pair that opens together.
• There’s no Face ID: a power-button fingerprint sensor adds a deliberate delay so users can read notifications before the home screen appears.
Why it matters: Founders must decide if the Duo’s premium user base justifies a layout redesign, as unoptimized apps currently look like legacy software on a $2,000 phone.
Worth it: Yes, for the specific UI failures and scaling behaviors that will define the foldable experience until developers catch up.
themobiletakeaway.com
• eToro ships its App Review reply as release notes @onmyway133 on X
Someone at eToro "accidentally copied their App Store rejection reply as Release Notes" for version v700.566.9, @onmyway133 spotted. The "What's New" text opens "Thank you for reviewing our submission" and answers Guideline 2.1 (Tap to Pay/ProximityReader) and 2.5.4. "And it still passed App Store review."
• Fortnite returns to Australia's App Store; rival iOS stores come in March @TimSweeneyEpic on X
Fortnite is back on Australia's App Store after a court ordered Apple to change its App Store rules in Epic's case. Tim Sweeney says the order opening up store competition takes effect in March, bringing the Epic Games Store and web app installs to iOS there.

Monetization

• Meta reports 10 million Muse setups and eventual ad integration Mobile Dev Memo
Meta’s Q3 2026 earnings show Muse reached 10 million setups, signaling a transition from its current subscription model toward integration with the company’s core advertising infrastructure.

Business

• Heyyyy and Whispy AI lead App Store revenue growth in September
ScreensDesign data shows Heyyyy's monthly revenue jumped 543% to $2.3M while Whispy AI reached $1.8M, marking a massive month for AI-driven chat and story apps.
Risers · August to September
Biggest revenue jumps across 54 tracked apps
Heyyyy: Stories that text back
chats with opinionated virtual personalities
+543%
$350k → $2.3M
Calmia: Health Tracker
estimates physiological signals using phone cameras
+400%
$40k → $200k
Whispy AI: Cozy Chat
provides personalized emotional support via chat
+338%
$400k → $1.8M
The Cringe: Chat Wrapped
analyzes chat history for communication insights
+267%
$15k → $55k
Sports Card Scanner: Spoly
identifies sports cards and tracks market value
+200%
$50k → $150k
Revenue estimates · ScreensDesign affiliate
Fireship artwork▶ YouTube · 7 min
Fireship
Key takeaway: Fireship’s satirical report on the Gemini 4 Argon release highlights the escalating AI benchmark war and new AI subscription tiers.
• Fireship describes Gemini 4 Argon as a native Omni-model trained on Google’s Ironwood TPUs with a March 2026 knowledge cutoff.
• Parodied benchmarks show Gemini 4 beating GPT-6 Astra in legal tasks (19.6% vs 5.4%) and agentic coding (77.9%) while losing in 3D rendering.
• The report mocks a pricing shift where Google locks new models behind a $99.99/month AI Ultra plan, excluding existing $20/month subscribers.
• Sponsor segment: Macroscope reviews agent-generated pull requests with Abstract Syntax Tree (AST) walkers that analyze code context rather than just text.
• It runs custom rules as native GitHub checks and auto-approves 40% of pull requests with a small blast radius.
• Macroscope claims 48.31% recall and review costs of about $1 per PR, versus $15–$25 for human reviews.
Why it matters: As coding agents increase the volume of pull requests, founders must adopt automated review tools to maintain quality without overwhelming human developers.
Worth it: Yes, for the commentary on shifting AI subscription tiers; the code review part is a sponsor read.
themobiletakeaway.com
Fireship artwork▶ YouTube · 5 min
Fireship
Key takeaway: Google DeepMind and University of Maryland researchers built Dream-RSI, a Recursive Self-Improvement (RSI) framework that lets AI agents optimize their own search strategies, drastically cutting the cost of solving complex engineering problems.
• Dream-RSI uses discovery logs to create a "replay simulator," letting an agent test thousands of search policies against previous runs without rerunning the underlying model.
• In a Lasso solver benchmark, Dream-RSI reached its goal in about 300 trials, versus about 51,000 for the previous record holder, SimpleTES.
• It runs at 1.79x lower cost than fixed exploration methods by caching every code attempt, score and crash status to disk for zero-execution-cost testing.
• Separately, researchers from ByteDance, Tsinghua University and others outlined a five-stage RSI roadmap, ending with a "Meta" stage where AI rewrites its own improvement logic.
• AlphaEvolve-style loops running thousands of autonomous evaluations are credited with high-level math results, such as a proof of the cycle double cover conjecture.
• Critics note the underlying model weights stay static; the AI isn't getting "smarter," just finding solutions faster through optimized search.
• 2026's breakthroughs come from these static models wrapped in orchestration harnesses and "swarms," not truly self-evolving intelligence.
Why it matters: As coding agents move from simple completions to autonomous problem-solving, RSI frameworks will determine which tools can fix complex bugs or optimize infrastructure at a commercially viable cost.
Worth it: Yes, for a technical look at how AI agents are moving beyond human-designed prompts to autonomous strategy optimization.
themobiletakeaway.com

Every issue