Mobile Dev Memo

Last April, in a piece titled Apple’s advertising ambitions, I posited that Apple’s introduction of view-through attribution to the AdServices API — as well as the renaming of Apple Search Ads to, simply, Apple Ads — could portend the introduction of ads to Apple Maps.

Strong unit economics make performance marketing scalable. Performance marketing creates good times. Good times invite growth hacks and viral gimmicks. Growth hacks and viral gimmicks weaken unit economics. Everything is an ad network 🫡

The fundamental issue with independent agentic commerce, and what almost certainly led OpenAI to shutter Instant Checkout after just a few months, and Walmart to pivot toward pushing Sparky into other chatbots rather than facilitating direct checkout, is what I've called Native Agent Proximity: If consumers demand to shop via agents with natural language, then on-platform agents will be the superior way to do that, and the largest eCommerce retailers already support that natively. Further, retail platforms have a number of structural, compelling reasons to reject working with independent commerce agents: ad revenue, promotional messaging, cross-selling, etc. A direct relationship with the consumer is invaluable. No large retail platform will give that up. But most importantly, the reason that 'agentic ad buying' never really took root, which relates to why the open web is declining as a destination for ad spend, is that independent agents simply perform worse than platform-owned agents. They have access to less data and therefore can't personalize as effectively. This isn’t a case of the price of anarchy; it’s the opposite: coordination is welfare-reducing. OpenAI shuttered Instant Checkout *quickly*, after just a few months. It wasn't because the concept is nascent; it's because the economics weren't tenable, Amazon wouldn't cooperate, and the outcomes were worse than native shopping experiences. These are fundamental roadblocks.

Private credit and the AI value reallocation As private credit markets deteriorate, it’s tempting not only to blame AI for that decline but also to extrapolate any dislocation to its logical extreme: that is, where rising default expectations among software companies are increasingly framed as the early signs of a global, systemic crisis. The most convenient analogy is the global financial crisis. But this framing is incomplete. It isolates the destructive effects of AI while ignoring the mechanisms through which those effects propagate value. In my piece this week, I argue that any weakening in various categories of the software landscape as a result of AI will not only mostly remain contained there but will likely lead to economically expansionary productivity gains and efficiencies that offset, potentially disproportionately, losses in private credit. mobiledevmemo.com/private-credit…

Might Meta step back from foundation model development? mobiledevmemo.com/might-meta-ste…

Apple has reduced its iOS commission in China ahead of World Consumer Rights Day on March 15th, when Chinese regulators investigate the business practices of foreign firms. As part of the change, Apple is reducing its App Store commission from 30% to 25%. But that's misleading: Apple has historically generated very little in App Store commission revenue in China outside of the largest mobile games (eg., Honor of Kings), which are directly distributed through the App Store. In 2025, Apple published an ecosystem study that cited $23BN in gross digital commerce on iOS in China for 2024, and I've seen that number cited in discussions around this commission fee reduction. But that number needs to be significantly adjusted: it accounts for *all* digital transactions on iOS in China, regardless of whether Apple could apply its commission or not (link-out, mini-apps, reader apps, etc.). So that $23BN “gross digital commerce” figure may have represented only $1.5–$3BN in commission revenue, or roughly 1.6%–3.2% of Apple’s total services revenue. That's because OTT super-apps like WeChat dominate the Chinese market. Until 2025, these apps paid no commission to Apple on transactions generated through the mini-apps they host, including mini-games. In November 2025, Apple reached an agreement to take a 15% commission from WeChat on those transactions. As part of the broader App Store commission reduction policy, Apple will now take a 12% commission on those transactions. So the comparison of this new 25% commission in China to what is paid on revenues generated in the United States is flawed. On a weighted basis, the difference is actually much larger: 30% for non-Small Business Program developers in the United States versus 12% for developers distributing through WeChat in China and 25% for large game developers distributing through the App Store (e.g., Tencent with Honor of Kings).

This week on the Mobile Dev Memo podcast, I interview Criteo’s CEO about the company’s RecSys system for retail use cases, its Criteo GO automation platform, and its recently-announced partnership with OpenAI. mobiledevmemo.com/podcast-recsys…

Platforms and native agent proximity One can’t consider the future of agentic commerce, or agentic advertising, as a competition between nimble upstarts and inflexible, staid incumbents: the largest platforms have embraced agentic interaction models. If you believe that consumers will demand to shop via prompts, or have their agents surface relevant items automatically on their behalf, or that advertisers want their entire workflow to be automated, then the largest platforms agree with you. mobiledevmemo.com/platforms-and-…

Netflix introduces a CAPI mobiledevmemo.com/netflix-introd…

A few weeks ago, I wrote about Meta's investments into GEM and its LLaTTE architecture, and last week, I wrote about Meta's use of LLMs for ad re-ranking (prioritizing the ad shown from a candidate set). Both of these approaches are attractive for ad optimization because they benefit from scaling laws: performance scales predictably with compute. And Meta's improvements in ad retrieval and ranking are generating meaningful revenue growth for the company. But these are large models, and scaling laws work bidirectionally. Two pertinent questions to consider in projecting that growth for Meta or for any of the large ad platforms: - What happens if the company becomes meaningfully compute-constrained? - What happens if the company has access to significantly more compute than competitors?

Agentic commerce is a mirage.

Google updates Google Play billing policy (and settles with Epic) mobiledevmemo.com/google-updates…

Can LLMs be used for ads ranking? Meta's GR2 architecture utilizes reinforcement learning and LLM world knowledge to improve ad ranking performance through structured reasoning. mobiledevmemo.com/can-llms-be-us…

Apropos of the Amazon-OpenAI partnership announced last week, I'm resurfacing what I published a few weeks ago about why Amazon would invest in OpenAI: to connect user accounts and source in-chat product discovery as an ad, not agentic commerce. mobiledevmemo.com/amazon-openai-…

AI Tracking Transparency In November, Apple updated its App Store developer guidelines to require developers to obtain affirmative, opt-in consent from users before sending “personal data” to third-party AI services. A pertinent question is: if Apple is able to dissuade developers from utilizing third-party AI services through the consent requirement, but accessing models hosted via PCC does not require consent, will Apple be able to charge a revenue share for “default AI status” at some point, as it does with default search engine status in Safari? mobiledevmemo.com/ai-tracking-tr…

For the past few months, I have been writing a long-form podcast articulating an AI Bull Thesis for the digital economy. I call it The Prosperous Society. In light of the Citrini report, I decided to chop it up and release it episodically. Episode 1 is live. mobiledevmemo.com/the-prosperous…

What is Meta’s GEM? mobiledevmemo.com/what-is-metas-…

AppLovin Q4 2025 earnings: understanding AppLovin’s risks mobiledevmemo.com/applovin-q4-20…