When Apple Intelligence (AI) debuted at WWDC ‘24 about a month ago, it spurred a debate about what AI is supposed to look like. If you think about it, generative AI is a huge leap in what virtual assistants have been trying to do for decades – provide the consumer with a tool that can make life easier simply by answering questions. The only difference is that these assistants have access to personal data that genAI doesn’t, at least out of the box. In fact, this is exactly what we envisioned AI to look like as well, be it through Hollywood or sci-fi books.
The reason why November 2022 changed the world is because the general public was provided access to large language models (LLMs). In simpler words, while assistants are limited to simpler language and mainly action-based responses, LLMs are much more advanced and can perform a variety of tasks. They excel at holding conversations and generating new, creative text across subjects. In fact, Siri is now going to utilise LLMs to provide smarter responses starting this fall.
The real question is – is AI a product or a feature? In other words, is it just an advanced virtual assistant or something that holds unique value deserving its own place in the digital ecosystem?
MKBHD answers this with a very clever approach of taking Clubhouse and TikTok as innovative examples. What Clubhouse offered was a live stage for content creators and celebrities to host their own audio-only stages for real-time drop-in conversations. It blew up during the pandemic. This idea was incorporated into Twitter as Spaces. And over time, it lost its charm. That’s an example of a product that offered an idea which was adapted by other larger players in the market.
Now, let’s look at TikTok – the largest culprit of lower attention spans in the modern world. Their idea was simple – create a platform where users could share and view short, engaging videos that capitalize on quick content consumption and viral trends. Creating tight and rapid feedback loops was the key to their success. This idea was also adapted by YouTube and Instagram into their platforms as Shorts and Reels respectively. But TikTok still holds its place in the market. It’s an example of a product that’s successful.
I find the success of AI to resemble that of password managers. Despite all major companies incorporating a version of it in their own operating systems, there’s always going to a place for 1Password and Bitwarden, simply because of how sophisticated they are. Apple Keychain is primitive when compared to 1Password, with lack of a standalone app (it’s coming this fall), nothing like Universal Autofill across the OS, and most importantly, no enterprise solutions.
Similarly, while Apple might’ve turned the tables with Apple Intelligence (AI) for personal use and deep integrations with the operating system, the need for general models from OpenAI, Anthropic, and Perplexity will remain. To be clear, the products are the ChatGPT and Claude websites, and the feature is Apple Intelligence or the Google Pixel’s Magic Eraser. The majority of OpenAI’s success comes from its enterprise solutions – such as the Copilot integration with Microsoft 3651 – which explains the sophistication these companies have, as I mentioned earlier. The bar to become big in this industry is also quite high - it takes a ton of money to build such models and then train them on expensive GPUs.
Interestingly, there’s a noticeable shift to feature-based AI for consumers, particularly personal AI. After all, we all want digital assistants that are knowledgable while being private and secure, a vision Apple demonstrated last month. As innovations in smartphones become less about the number of camera megapixels and more about software enhancements, and smarter personal assistants are a huge step in that direction, provided they check the boxes of being safe, private, and on-device.
💡 This blog post was inspired by MKBHD’s video – AI the Product vs AI the Feature