Profoundly Proactive

Two things can be true at once: newfangled AI personal assistants can be insanely powerful, and yet most people still have absolutely no idea what to do with them.

This is the inherent tension I feel in the discourse around the latest wave of AI capabilities currently sweeping over the land – well, at least the US (more on this in a minute). I hear from people – both online and in person – who have read my mostly positive writing about these new tools and still seem unsure what to make of all this. Is it just a fad or the next great leap in consumer technology?

One generally positive sign for it not being a fad is just how much everyone wants to talk about the trend – from people in Ohio (where I grew up) to The Economist (where I gave some thoughts for an article last week). Yes, that happens with fads too, but fads fade. Meta's Muse shows little sign of letting up as it's still atop the App Store now six weeks after launch. For as fun as Sora was, it didn't have this staying power. You know what has? ChatGPT.

Speaking of, while OpenAI chose not to launch their own stand-alone app for Dots, you can bet that if they did, it would be right there alongside Muse right now. Even GrokBot is moving up those charts. Copilot isn't far behind. And now Anthropic is moving fast to answer the proactive assistant call with Duet...

My bet would be that this is not all some flash in the pan but is instead the start of a real change in the way we live and work, both online and offline. And while the digital graveyard is filled with the bodies of AI personal assistants past, the difference is the actual technology here. LLMs, augmented by real-world training data, are now proving capable of handling such tasks – our tasks!

Still, capabilities don't always equal killer products – just ask Segway. The key here, for me, is the proactive nature of these services. Per my quote in said Economist article:

The big shift from earlier generations of digital assistants, explains M.G. Siegler, an investor and blogger, is that these new bots can work proactively, rather than reacting only to what users explicitly ask for. Instinct “really just works in the way I believe we all want AI to act”, says Noah Shinn, the startup’s 24-year-old founder.

Instinct, in a way, kicked this latest wave off. Sure, OpenClaw gets a lot of credit too, but again, beyond the capabilities, it's the proactive interaction that is going to be the key to this working, this time. And Instinct seemingly recognized this out of the gate, actively pinging users from the start to let them know what the service can do for them.

This is the jump start that many people need to get going – overcoming the "blank page" problem of not knowing where to begin – but just as importantly, it's also the key to keeping them going. A lot of people will come to these services with one idea to try, perhaps something a friend told them they were doing with such a service, but once that's done, they're often left with little more than a demo. And unlike the Alexa and Siri of old, thanks to the LLMs that now underpin these services, you can literally ask anything, which is a problem in and of itself – more of the "boil the ocean" variety. If you can ask for anything, you may be overwhelmed and not ask for anything.

Again, the proactivity is key here. Muse's 'Feed' is one clever way to get around these cold start (and cold continuation) issues, but the service more explicitly nudging you on certain things is far more powerful.

Obviously, these services have to be careful. If they're too proactive, it will likely annoy users and push them to turn the service off. Or if they're too "creepy" with their suggestions based on what they know about you, it will also likely turn people away. But there is a balance here, and in my early testing of both Muse and Dots, I think they straddle the line well.

But there's another key thing that people need to overcome in making all this work and that is being comfortable hooking in your email, calendars, reminders, etc., into these services. A lot of ink has already been spilt around Meta's trust issues (which is exactly why they doubled and then tripled down on security features at launch), and startups like Instinct and OpenAI (insofar as a $1T+ company can be a startup) will inherently lack such trust from users.

This is why I found it crazy that Google wasn't the first mover here. You don't have to trust Google in linking up all this data because they likely already have it! Apple is in a similar, and arguably better, camp thanks to the iPhone, and it's one of the reasons they're now seemingly situated pretty well in AI despite myriad stumbles.

Anyway, trust is an ongoing question in the space. And one breach of it could mean game over for any single service. But assuming that hurdle can be overcome, it's pretty wild what these services are capable of. They monitor your inbox and calendar and to-dos and figure out what they should nudge you about – and in some cases suggest actions they may take on your behalf. Permission is key, but you can already clearly see a future where certain tasks are trusted and done automatically in the background. Many early adopters are already doing such things – rather, they're having their agents do such things.

This is likely the future of email. The output won't be yet more email – which has long been the compounding problem at full internet scale – but it will be an actual to-do list, curated by your AI assistant.

There isn't a day that goes by now where I'm not sort of amazed by the proactive pings I get from one of these agents. The interactions range from as simple as reminding me to respond to something by a stated deadline in a message to finding linked emails related to a question in a thread in a way that would take me many minutes to do – if I could at all given the state of Gmail search.

That's one key to understanding all of this. It's not that what these AI agents are doing is impossible for me to do, it's that it's often tedious, to the point where I put it off as long as possible – often too long, and sometimes until it's too late. Even if you're the most on-top-of-it person, you're never going to be on top of everything and these assistants can actually help you save time with many tasks. I'm seeing it every day. Even if it's just 5 minutes per task, this time adds up quickly in aggregate. And it's time literally saved in your limited time of life.

Is that worth the money you have to spend on some of these tools? That's going to be a different equation for everyone. But there's also going to be a line for everyone. And I suspect that at least one of these services will cross it for pretty much everyone at some point. Even if it's just the built-in Siri AI capabilities of the iPhone.

I'm just a few weeks into daily usage here and my habits are already changing, fast. I'm now constantly getting pinged about new messages in my inbox by Muse or Dots ahead of me actually reading them. And they're far better about telling me what I need to know, or do, or watch out for in those emails. They're like a guardian angel of my inbox – it's sort of great. And it actually has me checking email less often, which is amazing.

Yes, that takes some level of the aforementioned trust, but it's slowly being established with every "aha" moment these services are proactively pinging me about. And it obviously goes far beyond the inbox.

Tracking pre-sale tickets for Dune 3, tracking iPhone Duo inventory, proactively creating reminders that I don't have to create myself, suggesting ideas for things to do while traveling based on previous calendar entries, sending me summaries of sporting events I missed due to conflicts in my calendar – the list literally goes on, growing daily.

To some, this will sound a bit exhausting. But again, the key is the timely pings about certain things – yes, just like a human assistant would do. But 99.9% of people don't have a human assistant, and this assistant happens to be one that may actually know more about you thanks to said service connections paired with essentially infinite memory and time to do tasks. It will take some getting used to – in a weird way, it's more about letting go of the way you used to do things, which will be the hardest part. But people will slowly get used to these new ways, and then they won't be able to imagine how they functioned without such things.

Bee, the AI wearable that Amazon acquired last year, is another tangential example of all this. When you turn it on, it just sits there and listens to any conversation and when it's over, gives you a summary of what was said and any action items required. It's honestly wild how well it works with just a single button press. And that simplicity and power could lead to a backlash because it's not clear the world is yet ready for this on a mass scale. And such functionality is about to come to the Apple Watch. But time and tweaks will alleviate those issues too.

I'm not suggesting that all of this will take over the world overnight – and hopefully not literally. For one thing, almost all of these services are US-only at the moment and they could very well be EU-never given the regulations there. Some will cheer this as they'll swear they don't want AI in their lives in this way. But even the story of the actual Luddites was less about the technology and more about the changes it wrought. There are obviously parallels here with the current AI backlash. The key to overcoming that starts with products that prove useful to people. With their proactive nature, personal AI assistants finally feel ready.