In June, I wrote about Agent Wars - how companies will build their own agents and fight over demand and context. Right now, personal AI agents like Instinct and Muse are the hottest AI category where this fight is happening.
OpenClaw and Hermes have been around for a while. But setting them up was a hassle. The turning point seems to be removing that friction and promising a more secure experience.
I’m waiting for access to Muse, but reading more about it led me to a few questions.
Will apps pay AI assistants for leads?
Muse lets me choose which apps it can access. But I can still ask it to compare options and recommend what suits me best. If assistants start influencing more of our spending this way, will their providers earn mainly from subscriptions and usage, or also from the businesses they recommend? (assuming the apps want to cede control over demand to these agents)
Suppose I connect two food-delivery apps and ask my assistant, “Order me a healthy dinner under ₹400, delivered before 8 PM.” I have chosen the apps it can use, but I’m asking it to recommend where the order should go.
Will businesses pay the assistant provider for leads or preferred placement? Are they competing to give me a better deal, paying my assistant’s provider more to win the order, or doing both?
Is the assistant still working for me?
A commission doesn’t automatically mean I get a bad deal. A business could give me the best offer and pay my AI agent’s provider. But what happens when another business pays more while offering something worse for me?
Would my assistant still choose what suits me best? And would I know if the payment influenced its choice?
I would want to understand whose interests determine its decisions.
The assistant’s value also comes from how much it knows about me. My food preferences, my schedule, how I work and the compromises I’m willing to make. The better it understands these things, the more useful it becomes.
But that leaves me with another question: what happens to everything my AI assistant knows about me? It could know more about me than any individual app. Does this become a profile that businesses use to target me or send offers to my agent?
Meta says Muse runs in a secure VM. But its own documentation says my conversations and records of the agent’s work can be used to train its models, unless I opt out. Keeping my files and profile in a VM is not the same as keeping all the context generated from them to myself.
I want an AI assistant that understands me. I don’t want my preferences and workflows becoming someone else’s training data or a profile to target ads. So how much of this needs to run under my control—not just the agent, but the models processing my data too?
Is this where personal private clouds and local models start to make sense?
Interesting read. The biggest question is trust — if AI agents start recommending what we buy, we should know whether they’re choosing what’s best for us or what benefits the platform more.