Hello again and welcome back to Fast Company is online.
For a long time, I used Siri for only two purposes. One was setting alarms, which he did perfectly. The other was answering random questions about the world that popped into my head. I asked those mainly because I was curious to see if I could answer them. Their success rate was around 50%.
But recently, I’ve been relying on Apple’s AI assistant as if it were, you know, an assistant. That was only possible with Siri AI, which became available this week as part of the public beta versions of Apple’s operating systems, scheduled for official release in the fall. The long-delayed realization of functions that the company announced more than two years ago, the new Siri incorporates Google’s Gemini model as one of its ingredients.
I keep asking questions partly because I wonder how Siri will respond. But now they are real questions whose answers will help me in everyday situations, and Siri is surpassing them. At a conference earlier this week, I was walking to a hotel for a lunchtime event when I realized I wasn’t sure what floor I was on. So I asked Siri. When I arrived at the entrance, I had an answer, no need to rummage through my calendar.
I don’t mean to say that it was in any way a technological advance. Other AI bots can use integrations to leverage external data sources, such as calendar apps. Everyone is smart enough to realize that my lunch location could be stored in my planner and figure it out for me. In my experience, they sometimes do work like this faster than the new Siri, which can be slow enough to test my patience. (The real test will be how agile it will be once it ships this fall.)
But since I’ve used Siri AI and tested similar tasks on some of its competitors for comparison, I’ve come to the conclusion that the basics of AI assistance are becoming a commodity. The differences between products will come in usability: how comprehensive, accessible and simply enjoyable it is to spend time with a product. On that front, everyone involved has a lot of work left to do.
Now that Siri is more like the ChatGPTs, Geminis, and Claudes of the world, it’s tempting to judge it by the standards they set. If you do, you might well conclude that you are still in recovery mode. Compared to almost anything else, it feels stripped down. Their answers tend to be better. He speaks with a voice that sounds like a human, but a human who is all business, not Scarlett Johansson. When I tested his funny side by asking him to tell me a story about a bear who opened a bakery, his story was so generic and dry that I could practically feel myself rolling my eyes. (When asked about their own baked bear stories, ChatGPT and Claude said they really enjoyed spinning a story.)
However, the more time I spent with the new Siri, the more its lack of interest in charming me seemed like a virtue. Other creators of AI assistants apparently consider chatterboxing to be central to the entire proposition, hence the term “chatbot.” More naturalistic voices, like those offered by OpenAI’s new GPT-Live, are an important focus for the industry.
But the things other AI assistants do to seem human and ingratiating (superfluous filler words like “ah,” manufactured enthusiasm, the incessant attempts to sugarcoat me) do nothing to improve the quality of the information being delivered. The busier I am, the more I prefer facts to be presented succinctly and blandly.
It turns out that Siri’s no-nonsense vibe is a design choice. As Apple software chief Craig Federighi explained in the podcast Mostly human:
[I]If you use a lot of the existing chatbots, they actually focus heavily on engagement. And adulation, right? In a way they want to attract you. They may encourage you to reveal things about yourself and then use them as a basis for establishing a connection.
We see it quite the opposite. I mean, the way we’ve designed Siri, Siri really wants to say, “Listen, I’m not here for that, right? I’m here to help you. I can help you do things. I can help you learn about the world.” But if you try to involve Siri as a romantic partner, Siri isn’t ready for that.
(Disclaimer: I did not try to involve Siri as a romantic partner.)
Meanwhile, OpenAI’s GPT-Live, which the company is promoting with a video showing several endearing old ladies talking confidently, is fundamentally dumber than previous voice versions of ChatGPT, at least for now. It doesn’t support the connectors and add-ons needed to do jobs like checking email and calendars. It also lost the ability to interpret live video from a phone’s camera, a feature OpenAI first demonstrated more than two years ago. There is a lot of substance to lose in something that is theoretically an improvement.
Siri is also a truly plug-and-play experience, in a way that makes other assistants seem like even more science projects than they already were. It doesn’t expect you to choose different models depending on how demanding your question is. Nor do their abilities seem to vary between spoken and written modes in any way that is not obvious. (Like GPT-Live, Claude Voice mode doesn’t support integrations, and neither GPT-Live nor Claude Voice are self-aware enough to explain their own limitations or point out that you can work around them by typing.)
I now recognize that Apple has some unfair advantages on its own platforms. Only Siri is integrated into Spotlight Search and Dynamic Island, and can be invoked with “Hey Siri” or by long-pressing the iPhone’s side button (although you can still ask Siri to pass directions to ChatGPT, and you can program the action button to open any assistant you want). But the company deserves credit for considering the integration of Siri AI, which seamlessly transitions between voice and display modes depending on how you interact with it. Wherever Siri goes in the future, I hope Apple doesn’t alter this basic approach.
I also hope that Siri finally becoming respectable doesn’t lead developers of other assistants to give up on creating ambitious versions for Apple platforms, like they were Windows browser developers from the ’90s who concluded they could never compete with Internet Explorer. Even though I find Siri useful, I still use Gemini, Claude, and ChatGPT a lot, as they all have their strengths. The world needs AI assistants with priorities that stray from Apple’s, all the way to nerdy and potentially dangerous powerhouses like OpenClaw. I’m looking forward to seeing you all evolve.
As I wrote last month, Apple managed to avoid serious damage to its AI position despite the two-year delay of Siri AI. In part, that’s because many of Apple’s rivals spent some or all of that time trying to find their own footing. For example, Microsoft hired a DeepMind co-founder, Mustafa Suleyman, to oversee a consumer version of Copilot before deciding that there shouldn’t be a separate consumer version of Copilot after all. He ended up redeploying Suleyman to focus on model development.
Even OpenAI is having second thoughts about how ChatGPT should work. Last week, it combined the assistant along with its Codex encoding agent and the Atlas browser to create an all-in-one app, a move that has been poorly received.
Speaking of OpenAI, on Monday, Bloomberg’s Mark Gurman gave the scoop on the hardware product he’s been working on with former Apple design legend Jony Ive. According to Gurman, it is an ambitious AI companion in the form of a battery-powered portable speaker without a screen. It reports that OpenAI may announce the product this year and ship it in 2027 (although Apple suing the company for trade secret theft may not help).
I’m not going to form any opinions on OpenAI’s device and its chances of success until we know more than a few superficial details about it. But the worst-case scenario would be for OpenAI to be distracted by another side mission when there are still so many opportunities to improve its core product. In the years to come, no AI assistant will matter more than a truly great one that runs on smartphones. And the opportunity to create it is still open.
More top tech stories from Fast Company
Why Apple cares so much about metal finish
In Apple’s lawsuit against OpenAI for theft of trade secrets, more than a pretty metal finish is at stake: it’s the entire innovation process. Read more →
Satya Nadella defends the independence of AI
Microsoft’s CEO noted that companies should control more of the data, models and infrastructure that make AI useful. Read more →
Waymo’s July 4 chaos in San Francisco raises new questions about how robotaxis can work at scale
After their vehicles stalled and worsened traffic jams near the fireworks celebration, city officials say they still lack the tools and data needed to manage autonomous fleets during major events. Read more →
The FCC just approved the testing of a giant mirror in space
Scientists warn that Reflect Orbital’s plan to send sunlight to Earth after dark could disrupt the sleep of wildlife and humans. Read more →
How an army of digital detectives is using AI to fix America’s crumbling sidewalks
Project Sidewalk combines machine learning and community engagement to map and improve sidewalk accessibility. Read more →
AI economics is based on this (incredibly vague) unit
Users are still getting used to the token. Read more →
have you been reading Connected, fast companyThe weekly technology newsletter from me, global technology editor Harry McCracken. If a friend or colleague sent you this issue (or if you’re reading it on fastcompany.com), you can check out previous issues and sign up to receive it yourself every Friday morning. I love hearing from you – ping me at hmccracken@fastcompany.com with your comments and ideas for future newsletters. I’m also into Bluesky, Mastodon and Threads, and you can continue Connected on Flipboard.