- Road to Artificia
- Posts
- In the era of AI agents, Apple keeps agency for itself
In the era of AI agents, Apple keeps agency for itself
How do agents fit into the Apple Intelligence strategy?
Jeff LaPorte
December 09, 2024 • Estimated Reading Time: 11 minutes
Did you receive this forwarded from a friend?
AI agents are the new apps. But with Apple Intelligence, devs are just data donors and task runners - locked out of the agent layer entirely.
The iPhone SDK Precedent
During the original development of the iPhone, Apple’s senior execs debated whether the new platform should be open to third party applications. Scott Forstall was the Apple SVP who led the creation of iOS prior to the launch of the iPhone. In his deposition in Epic vs Apple case filed by Epic in August 2020, he described the debate:
Forstall: There were executives at Apple that thought we should never release the ability for third parties to do any natively compiled applications.
There were executives who thought –– and they thought we should just have web applications and –– and then websites running with web standards inside of the browser or browsers on the platform, the browser.
There were executives who thought we should have a hybrid model of some web technologies and some native abilities. And then there were executives who thought we should provide a platform to enable third parties to build fully native applications on the platform.
Indeed, when the iPhone shipped in June 2007, it shipped with no support for third party applications. Steve Jobs tried to soften the blow at the iPhone announcement event by telling developers they would be able to create web apps, a “sweet solution”. Immediately after the iPhone’s release a community of jailbreakers formed, hacked the app install process, and started building native apps with reverse-engineered iOS APIs. Before the end of October, Apple had switched gears and announced that a native Software Development Kit (SDK) would arrive in a few months. Apple got to work building the App Store, and launched it to the world in July, 2008.
Why did Apple try to do something so “dumb” as to think they could create a new platform with no support for third-party apps?
Ben Thompson of Stratechery has written compellingly about Apple’s late 90’s near-death trauma, which he found etched into the company’s culture during his time there:
Ben Thompson: I had a unique vantage point into how Apple looks at this time in its history due to my time spent on the Apple University team, and something that struck me was the powerful impact Apple’s near-death experience had on all those involved. That Wired cover, and the fact Apple lost nearly a billion dollars in 1996, came up unprompted again and again. For those that were there, it still fundamentally shapes how they view the world and how they make decisions. (emphasis added)
It seems almost absurd now as they sit atop the market with a value of $3.6T, but Apple’s experience of a time when they were at the mercy of app developers like Adobe and Microsoft etched itself into the company.
This history of resistance to third-party developers is relevant again, as AI agents emerge as the most important area for innovation. Indeed, Apple's approach to Apple Intelligence reveals similar tensions between control and innovation.
Apple Intelligence
Fast forward to this year. Apple Intelligence is announced in June 2024, with capabilities being gradually released to customers since October. The user experience is not yet delivering on the most compelling use-cases from the announcement, but the direction is clear and what has been released is popular with users.
But, although the full capabilities haven’t landed yet, the architecture presented at this year’s Apple worldwide developer conference (WWDC24) is clearly defined.
The design shown below looks a bit complicated, but we only care about a few pieces (circled in yellow):
Apps: Your existing apps, enhanced to share data and functionality with the system
Siri: The user-facing interface and primary entry point for AI interactions
Semantic Index: A special kind of database, organized for easy retrieval and use of data by AI models. Your apps will contribute their most important pieces of data to the index.
an App Intents Toolbox: A collection of task types that can be issued by an AI model to your apps. It defines standardized ways to perform actions like "schedule meeting" or "send message," allowing AI models to use features of your apps.
AI Models: A collection of on-device and cloud-based language models that power understanding and decision-making.
Source: Apple WWDC2024
The system involves a set of optimized models, living on both your device and in the cloud. Apps can expose their data (for instance, a calendar event or a group chat) to an Index for later retrieval by an language model. And apps can expose their functionality to the system, and respond to requests to perform actions (“Intents”, explained below) using that functionality.
And on top of it all sits Siri.