It feels like the product layer is getting the least attention right now. Some attention has moved from models to the harness, and now a little more flowing to agent identity. Chat is such a tired way of interacting with AI, which are capable of generating both UIs and visuals on the fly, and understanding audio and video.
Think Claude Code with their focus on HTML docs and Artifacts has broken new ground. OpenAI followed up with ChatGPT Space last week. Apps like https://codoc.sh are interesting experiments, and hopefully there will be more multiplayer at the product layer soon.
Just like how agents wrapped models (e.g. when using Instinct you don't care which model it's using, and very likely it's using many different ones depending on the task), I suspect soon you won't care which agent you're using - I mean, you might decide between Muse or Instinct or Grok Bot, but you'll have multiple of them, and you might have a "family agent" that you share with your wife (to use an example from your post), and you might be able to directly talk to your wife's agent (e.g. in a group chat or a shared travel plan doc). The separation will be along the lines of: which agent has what permissions, what overarching goals you've set, whether shared with other people.
It feels like the product layer is getting the least attention right now. Some attention has moved from models to the harness, and now a little more flowing to agent identity. Chat is such a tired way of interacting with AI, which are capable of generating both UIs and visuals on the fly, and understanding audio and video.
Think Claude Code with their focus on HTML docs and Artifacts has broken new ground. OpenAI followed up with ChatGPT Space last week. Apps like https://codoc.sh are interesting experiments, and hopefully there will be more multiplayer at the product layer soon.
Just like how agents wrapped models (e.g. when using Instinct you don't care which model it's using, and very likely it's using many different ones depending on the task), I suspect soon you won't care which agent you're using - I mean, you might decide between Muse or Instinct or Grok Bot, but you'll have multiple of them, and you might have a "family agent" that you share with your wife (to use an example from your post), and you might be able to directly talk to your wife's agent (e.g. in a group chat or a shared travel plan doc). The separation will be along the lines of: which agent has what permissions, what overarching goals you've set, whether shared with other people.
You may be right. It's all so fluid right now, hard to believe that OpenClaw burst on the scene just a year ago!