9to5Mac brought to you by Memoket: Memoket Gem is a 0.4 oz AI wearable that sits next to your Apple Watch. A single press captures a conversation and KITE memory links it to what you said weeks ago. 20 hour battery. Available now starting at $168.99 memoket.ai And Amazon.
While it’s great that Siri AI can finally deliver on Apple’s promise of a truly useful personal assistant in the age of AI, expectations for assistant agents are rapidly evolving. But Apple is showing that it may no longer watch from the sidelines.
Siri AI is great, but it’s also very 2024
Now that Siri AI (which is still officially designated as beta) has been available for a little over a month, it seems fair to say that its two defining characteristics are that it’s finally useful, but it can be painfully slow to respond to even simple queries.
And while I’ve spent much of the last few months mocking Silicon Valley’s obsession with AI assistants that can book flights and hotels, I hate to admit that the first time Siri AI truly impressed me was when I asked how much I paid for a hotel room for an upcoming trip to San Francisco and how far away my first date was in the city. Both answers were completely correct.
Could I have looked for it myself? Of course. But aside from the fact that I was in the middle of a 10K race at the time, finding this information on my own probably would have taken longer than it took Siri to give me the answers.
As a result, over the past few months, I’ve relied more and more on Siri AI for questions like “What were the contact details for that plumber who fixed the kitchen sink leak a few years ago?” or “What was the name of the PR person who introduced me to so-and-so?”
Information that I know I have somewhere, whether it’s an iMessage conversation or an email receipt, and that even if I could just look it up, I now prefer to have Siri AI do it for me.
And for the first time in 15 years, I can ask these questions without expecting Siri to fail halfway through asking them, only to be frustrated when she inevitably gets it wrong and proves me right.
This is all very good. The problem is that Apple is only just catching up on what AI assistants could do in 2024.
Apple’s next AI challenge: creating an agent harness for Siri
Over the past few months in particular, the AI frontier market has increasingly shifted toward agent assistants, with products from OpenAI, Anthropic, Meta, and xAI (among many others) going beyond answering questions to actually performing tasks on users’ behalf.
I subscribe to ChatGPT and use Codex and ChatGPT Work extensively for everyday tasks involving files and information scattered across my Mac and various cloud services.
And it always seems magical whenever I ask one of these tools to do something that involves opening, creating, editing, or organizing files on my computer, as well as interacting with websites and services I’m connected to, and it succeeds.
The flip side is that every time ChatGPT performs one of these tasks, the new Siri AI starts looking like the old Siri again.
It’s still early, of course, but this already appears to be another area where Apple appears to be falling further and further behind, as AI labs continue to improve their agent capabilities, sometimes rolling out major updates every few weeks, while Apple remains largely tied to its annual software release cycle.
But there is a significant limitation to today’s agent assistants and I think Apple is uniquely positioned to address it.
As impressive as tools like Codex, Dots, and ChatGPT Work can be, they still rely heavily on the user to provide permissions and context behind the task they are supposed to accomplish. They can access files, search emails, and interact with websites, but they usually don’t know what you’re doing, what you’ve discussed with other people, or what you might need next unless you explicitly tell them.
And as these tools become more and more capable of operating our computers and phones on our behalf, I suspect that the next big step forward in their usefulness will come not only from improving what they can do, but also from improving their proactive understanding of the user’s context in real time.
This is where Apple might have an advantage.
Apple recently showed it’s finally ready to capitalize on the trust it’s spent years building around privacy, even if it means testing the limits of what users are comfortable sharing with an AI assistant.
Last month, it introduced two Apple Watch features: Siri Recap, which can automatically generate summaries of conversations throughout the day, and Live Rewind, which lets users catch the last 15 seconds of a conversation they may have missed.
Both features rely on the Apple Watch’s microphone to capture information about what’s happening around the user, potentially giving Siri access to a whole new source of personal context.
Of course, the privacy implications are enormous, not least because the conversations involve other people who have not necessarily agreed to have their words processed by an AI assistant.
And while these concerns are certainly worth discussing in detail (and likely explain why these features haven’t been rolled out yet), Apple has repeatedly shown that it is willing to go to extraordinary lengths to address the privacy challenges associated with technological access to sensitive personal information. Just look at the efforts put into developing its reference image system.
As assistant agents become more capable, I suspect their biggest limitation will have less and less to do with what the underlying models can do, and more to do with what we are willing to let them know about us.
Imagine an assistant that doesn’t just wait for you to ask it to find a document or complete a task, but already understands what you’re working on. One that can collect information in the background so it’s ready when you need it, or even take action on your behalf based on what’s happening around you.
It’s the kind of assistant that could make today’s agent tools as limited as the old Siri (and, increasingly, the new Siri AI) does today.
And while Apple still has a lot of catching up to do when it comes to developing these capabilities, it’s in a much better position to compete in this next phase of AI than it was at the start of the generative AI race.
9to5Mac is brought to you by Memoket:
Apple just added voice recording and note taking to the iPhone and Apple Watch. Useful, but check your battery at 2 p.m. Dedicated recording is one of the quickest ways to drain a watch, and your iPhone must be out and unlocked to do it. More importantly, neither device connects what was said in today’s meeting to a decision you made three weeks ago. They capture moments. They don’t build memory.
Memoket Gem was designed to do more. A 0.4 oz wearable AI agent that fits next to your Apple Watch on the same wrist. One button, one press, 20 hours of battery life. Gem captures conversations and links them over time using KITE, a long-term memory engine that connects what was said today to what mattered weeks ago. When a topic comes up, Gem automatically brings up the original conversation. Everything takes place in ChatGPT, Claude, Slack and Notion.
153 languages supported. Available now on memoket.ai And Amazon starting at $168.99.
FTC: We use automatic, revenue-generating affiliate links. More.
