Age of agents
We’ve all been looking at AI agents for a while now: probably since before OpenClaw, but it really accelerated this year after OpenClaw and its derivatives came out.
I’m not sure why it took nearly six to eight months after OpenClaw, but we’re finally seeing all sorts of personal AI agents come out: agents that you can just sign into and use out of the box. Maybe it’s just that the models truly are that good now. Maybe having small features like a bridge to iMessage did the trick to convince more users to try them.
Last week, it was Instinct – everyone you knew was talking about it non-stop. Today, it’s Muse from Meta. (The way this rolled out immediately after Instinct went public makes me wonder if Mark approached the company to drop his tried-and-true tactic: “Sell to us now or we’ll clone you next week.”
Instinct was fine. Like other off-the-shelf agents, it didn’t do that much more than our own family agent at home was already doing (We rolled our own, Robin – @roundrobin, a couple of months ago. More on that in another post.). But Muse is really good.
Muse has a lot of things going for it over all the others: just take a look at Assistant Benchmark. For all you want to hate on Facebook and the fact that they have fallen behind in models and being seen as the platform of choice for AI, never bet against them. Muse can:
- be free longer than Instinct (and others) can probably stay alive before needing to raise another round of financing,
- use all of Meta’s compute capabilities,
- monetize with ads (already a core strength of the company),
- monetize with commerce (see Collison’s tweet about Link integration),
- use Meta’s experience on how to distribute software and to cross-sell it across the other apps in their portfolio,
- seamlessly tie in data from other Meta apps you’ve been using for years (see my Instagram example in the picture).
The ads bit might be the most interesting: they already have the engine for it. The only way you’re going to get to a billion users is if the product is free*. Therefore, at some point, after you’re done spending your go-to market cash, you’re going to need ads or some other layer that makes money without the user directly paying for it.
A “pro”, besides the cost (free), might be that big players like Meta and Google are the only ones that can make an agent like this. You don’t want to sit around linking up all your accounts to a brand new startup. You have to deal with more points of friction, the biggest being their lack of knowledge and past context about you. The downside, of course, is that you have to choose which giant company you want handling all your data. Ultimately, it’s going to come down to whether you trust Facebook (or Google or Apple) with all your stuff and all the connectors to all your other stuff they don’t already have (e.g., Resy, OpenTable, airline logins, insurance, cars, Airbnbs, campgrounds, schools and anything else where you’d like an agent to log in on your behalf).
I’m looking forward to seeing what Apple will announce tomorrow and in the coming months on this front. Personally, I’ve split my “cloud” across a few platforms – email on one, photos on another, social (or whatever that means) on smaller networks, logins on another and so on. If there was any company I’d trust to have wider access to even more of my data with the intention of making my life easier with agents, it would be Apple.
* Or, say, 90% of the product is free to 90% of users; 10% can pay for some sort of premium access.