I’m using my imagination here, so bear with me.

At the moment, AI agents via LLMs are very input-heavy. A user asks ChatGPT to reword his email before sending it to his colleague. Or a user asks Gemini to create something from scratch.

What if the AI agents and LLMs were instead applied to the outputs? For example, I plug an LLM into my inbox so that every email that I read is re-written to be in an uplifting tone. Then it doesn’t matter what AI the sender used to re-write the text. Or I ask my AI to remix a photo from Instagram based on my preferences, and not on the preferences of the creator. Think of it like MyAI, where there is extreme personalization.