Context-Driven Roadmap
Yesterday, Google announced Personal Intelligence for Gemini - the next iteration in hyper-personalised AI. Personal Intelligence pulls in a user's data and context from across Google's suite of products to deliver more relevant AI responses within Gemini.
This was always the screamingly obvious product path for Google and arguably the most profound reason to have the heebie-jeebies if you're all in on Open AI (the others being Alphabet's/Google's gushers of free cash flow that makes it immune to VC funding markets, proprietary TPUs that makes it immune to NVIDIA's dominance, and not-great-but-at-least-out-there consumer hardware).
As the OG of the internet era, Google has an extraordinary depth of user information and can combine it across multiple products. Pre-AI, my Google Search data was a pretty good proxy for every area of 'intent' in my life. YouTube data is and continues to be a pretty good proxy for things I'm interested in. G-Suite (Gmail, Google Calendar, Google Drive, and so on) helps me organise my life - and for many, their work - so knows pretty much everything about who I'm communicating with, who I'm meeting, where I'm going, and so on.
Surfacing all of that as context to an AI is a superpower, and a classic example of a 'Cornered Resource' in the Hamilton Helmer 7 Powers framework (one of my strongest Rec's). As the (relatively) new kid on the block, Open AI has a major disadvantage. Sure, ChatGPT was a breakout product that had arguably a 3-year lead on the rest of the market when it comes to end users interacting with LLMs and the new format of Search they offer. And that's a decent amount of time and use case to amass a formidable amount of user data. But OpenAI remains, for now at least, largely a one trick pony - at least on the consumer side.
I imagine the need for additional user context to compete with hyper-personalised AI is one of the driving forces behind OpenAI's product and feature releases (another being the desperate hunt for network effects - a topic for another day). Sora - or maybe more accurately, Sora 2 - is trying to capture the same user behaviour and context as YouTube. The Atlas browser is an attempt to capture more data on what users are using the internet for outside the chat window. The mysterious io Projects acqui-hire of Jony Ive, regardless of the hardware device's ultimate form factor, is pointed towards AI living alongside us in the real world, experiencing what we experience. ChatGPT Health is a land grab for medical history and other health data.
Yes, all of these things have standalone value to Open AI, but surely a huge slug of the strategic rationale to dedicate resource to these areas is the ability to weave together a patchwork quilt of a user's multiplicities based on numerous contexts and use cases. And that driving imperative, I think, will continue to be a key plank of the strategic rationale for new product/feature releases, acquisitions and acqui-hires.
It also points towards a potential pocket of lower competition for consumer start-ups aiming to sit alongside and compete around the edges with the major AI platforms. The simple articulation of the idea is this: the lesser the value of the product context as part of a wider tapestry of user understanding, the less likely that product is to be on the Open AI 'start-up steamroller' aka product roadmap.
Isolated bets don't feel strategic, and in a world of limited resource - if not capital for now, at least compute and consumer attention - I'd wager these projects will have less appeal around the board table at Open AI HQ.