Agent Switching Costs and the Case for Open Memory

Agent switching costs are going to be immense.

Think about everything your agent will learn over time: your work, your routines, your preferences, who matters to you, how you like things done.

That context compounds with every conversation. And every day you embed an agent deeper into your life, leaving it gets more expensive.

Maybe we need a startup that serves as the secure repository of that information, and then you connect it to Muse, Grok Bot etc.

Or open standards for memory. So what your agent knows about you can move with you.

Open Memory. Your data is agent-agnostic. The agent is interchangeable.

I don't know how this is solved. Agent companies benefit enormously from lock-in. And an agent's memory isn't just portable data. A lot of what makes an agent feel like it knows you is how it decides what to remember and when to use it.

But I still feel like the thing an agent learns/knows about me should move with me, between agents.

Then you have experiences like Grok Bot being on your phone, computer and car. That's what you want. One agent that moves with you throughout your environments. But that seamlessness is exactly what locks you in.

It's way too early to know which agent will be the winner, given the pace of model releases and constant improvements. Who knows if the agent paradigm, in its current form, will be around in 12 months.

Which makes it even harder to decide which agent to go all-in on!