The Day Generative AI Became a SubFeature Rather Than A Product
Plus a short history of where these ideas originated
On June 10th 2024, Apple announced a series of new GenAI features at its 2024 WWDC they call Apple Intelligence. Private and secure actionable agents. A focus on productivity rather than cute tricks. And this also marked the day that it became clear to me that Generative AI is going to be a sub feature of features of apps on devices (and credit to Jessica Lessin who wrote that AI is now a feature rather than a product). At least as far as consumer usage is concerned. For example the iPhone has an Email app which has features like writing or sorting emails which will now have sub features powered by GenAI which will help you compose, sort, prioritize responses and so on. And furthermore, interact between applications as an agent on your behalf like “take that picture I took of the tree an hour ago and put it in an email to my mother saying how much the tree has grown and send it in an hour.” All powered by your voice. And this is totally amazing! Almost like science fiction. And it is the world we envisioned in many sci-fi movies. And they will also be as commonplace in our lives within a year as email or messaging itself.
A little personal history
Much of what we are seeing now was envisioned at a company called General Magic. In the mid 1990’s Magic created the concept of agents working on your behalf and even developed a programming language to enable them called Telescript. The agent vision was the brainchild of Marc Porat which include the idea of agents working in the background even for months on your behalf coordinating between live data on websites and your local apps. The company also developed hardware that at first looked quite like the iPhone of today.
Later, in the late 1990s at General Magic, we invented the first true virtual assistant, powered by your voice. I humbly led that “all-star” team which included Gary Lang (Microsoft, Amazon), Danny Lang (IBM, Unity, Google, who worked on VoiceXML), Glen Shires (Google), Viki Paige (Broadcom) and about 100 others. The platform was Portico, and her name was Mary. She could understand millions of phrases and had salient and interesting responses on the order of thousands. This seminal work resulted in a dozen patents and predated the founding of products like Siri, Tellme, Alexa and others. Those systems borrowed from the Portico inventions and technologies laid out in the patents. And in truth we also all knew each other because Silicon Valley is quite small. Eventually all the major players licensed the General Magic patents since one could not create a humanlike interaction or even send vectors to the network for speech analysis without running into these patents. Mary had relative context, meaning that she knew what you said or were doing last and that what you said next likely related to that. She could answer your phone for you, check that the caller was a known contact, open your calendar, search for openings, suggest some to the caller, and book them on your calendar, all while you were sleeping. She could read your email, respond to emails, read your calendar to you, get stock quotes, read web pages, weather, sports, traffic. Quite the consummate agent for 1998!
General Motors launched the OnStar Virtual Advisor to over 2M subscribers using the Portico system, and many telcos such as Qwest also launched branded versions leading to millions of users.
You can watch the teaser from 1997 of Portico (then called Serengeti) here, hear the 1999 advertisement here, and watch an on stage demo here.
At the time we saw this as a network service mostly because it was not possible that we could do all this processing and agent work on device at the time. The iPhone was still 9 years away from launch, and a flip phone had limited CPU resources. And an M1 processor was 20 years away. So, it was very network centric. So much so, we built our own ops center (there were no cloud providers yet!).
Yet the idea of voice driven agents continued to slowly march forward as the ultimate user interface to what you want to get done.
Back to 2024
Today we have processors, including the M1-M4 series from Apple that have neural engines onboard, able to handle many of these tasks in seconds without jumping out to the network. And this is what starts to make all of this become seamless subfeatures. So seamless we won’t even think about whether its AI or ML or rules or math or whatever. Won’t matter because we just want to get work done efficiently. And this is where the full vertical integration of Apple works well. They own the hardware, the processors, the OS, the cloud service and many of the apps. So what they can pull off as seamless no one else can.
In Apples announcement of features, they easily put 1000 startups out of business. So many AI startups built an app atop GPT4 that adds a little value. But all that value is being swept up by Apple integrating it into daily tasks in ways that seem perfect. “Siri, fix that photo by changing the background.” Maybe there are 25 companies who have some app to do that and now there are none. I won’t go out to any 3rd party app for this (that’s a product). It’s now just a subfeature within a feature (photo editing) within an existing app (Photos). And it already all works. That company that made an app for that….dead. I know it had a $5B valuation with no revenue…but VCs should know better. Its dead.
As for the foundational model makers (ie OpenAI) I see this quickly being commoditized as well. There are several models (Gemini, Llama 3, Mistral) that are closing the gap quickly between GPT4o and themselves. Eventually all will train on similar data and they will all seem the same. And that will drive down prices to the cost of the electricity to run the GPUs. Thus a commodity business and one that will get there quickly. Like 2025. Who Wins in that case? The existing providers who have other income streams. Maybe Microsoft ends up buying OpenAI…or replaces them with their own model.
Despite challenging times ahead for AI app startups and perhaps foundational model providers, we the users are getting amazing capabilities this year that we have dreamed of for 25 years. These are life-changing productivity boosters you’ll use hourly. And in a year you won’t even give it much thought.
