[scuttleslops] AAPL, NVDA, TYL, FISV, APO, ROL
Upcoming posts (in no particular order): Gartner, Visa/Mastercard, Tractor Supply, U-Haul, Apollo, Zoetis, and CCC
I sensed some skepticism when scuttleslops was announced. But now that readers have had a chance to see the product, the feedback I’ve gotten has been uniformly positive, so I plan to continue this project for the foreseeable future. A few things:
First, after publishing several scuttleslops, it’s become clear that the service still very much requires a human in the loop. While the text is still substantially AI-generated, I am more involved in the editorial process than I set out to be. These are “scuttleslops”, sure, but I still have quality standards and the AI output, unedited and taken at face value, fails to meet them. To my dismay, this puts more work on my already crowded plate; to my relief, it suggests I’m not yet completely replaceable!
Second, while scuttleslops was originally intended to summarize news items, I’m expanding service to include podcast interviews, sell-side interviews, and – sometimes, with permission – expert call interviews hosted by InPractise, a service that I’ve subscribed to for years and recommend to investors interested in insightful expert interviews that address the meaty, structural considerations that matter over years, not months.
Stratechery Interview with Ben Bajarin About Apple, AI, and Compute; June 11, 2026
The interview frames Apple’s WWDC announcements as less of a flashy AI product launch and more of a platform re-architecture required to make Apple Intelligence and Siri work at all. Bajarin argues that Apple had to modernize foundational layers like the CPU scheduler and, more importantly, the search index, because useful Siri is mostly “fancy search” across personal context, messages, photos, calendar, and applications.
Apple does not need to beat OpenAI or Anthropic at general-purpose frontier models if it can own the consumer control plane through the operating system. Bajarin thinks consumers do not really care about “agents” as a concept; they care about hiring Siri or another assistant to get a simple job done, such as finding a photo, summarizing an email, changing a calendar item, or answering a contextual query.
Thompson pushes this further, arguing that consumer and enterprise AI are fundamentally different markets because “consumers don’t want to do work,” while enterprises pay for productivity gains. That distinction is important for Apple’s moat: Apple may win consumer AI not by building the most autonomous agent, but by embedding useful, private, low-friction AI into the device experience, while OpenAI’s consumer ambitions become harder because it lacks system-level access to Apple Mail, Messages, Photos, and Calendar.
OpenAI may have a weaker path in consumer devices because Apple can satisfy many mainstream AI use cases at the OS level, while Google benefits by supplying the cloud/model layer behind the scenes. Bajarin even says he has “zero” faith that the OpenAI-Jony Ive device changes this, which underscores the view that consumer AI hardware without system-level context and distribution is a very hard market.
On devices, Thompson and Bajarin draw a distinction between answer inference and agentic inference. Edge devices, including Apple devices and future Windows AI PCs, may be useful for lower-cost “answer inference” such as summaries, contextual answers, and lightweight local generation, but Thompson is skeptical that complex agentic workflows will run locally because memory and orchestration requirements favor the cloud.
Bajarin partially agrees but argues that enterprises will still care about local token generation because it can reduce recurring cloud token spend, making high-end Macs or Nvidia-powered Windows PCs economically attractive if they save enough AI operating cost. This is where Nvidia’s new PC chip matters: its unified memory architecture gives Windows OEMs a way to compete with Apple’s edge-AI story at the high end.