Google dropped a new Gemini feature this week that lets you import chat history from ChatGPT and other chatbots. The AI data portability crowd is excited, and I understand why on a surface level — nobody likes feeling trapped in a platform. But I keep staring at this announcement and thinking about how thoroughly it misunderstands what “context” actually means for an AI that’s supposed to help you get work done.

Porting conversations is moving furniture between burning buildings

The pitch is straightforward: you’ve spent months training ChatGPT to understand your preferences, your writing style, your projects. Now you can bring all of that into Gemini so you do not start from scratch. And sure, that sounds nice in the same way that transferring your Spotify playlists to Apple Music sounds nice. But chat history is not institutional knowledge. It is a log of questions you once asked, many of which you have already forgotten about, and answers that may no longer be accurate because the models have changed three times since then.

I went back and looked at my own ChatGPT history from six months ago. Half of it was debugging code for a project I abandoned. A quarter was research for a blog post I already published. The remaining conversations were so context-dependent on what I was looking at on my screen at the time that importing them into another chatbot would be like handing someone your old grocery lists and expecting them to cook dinner.

The context that actually matters is not in any chat log

When I need AI help with work, the relevant context is almost never something I discussed in a previous conversation. It is the page I am currently looking at. The email I need to reply to. The form I’m filling out for the third time this week because some enterprise SaaS app refuses to remember my department code.

That context lives in browser session state — cookies, login tokens, open tabs, the DOM of the page you are staring at right now. No amount of chat history import captures any of this, and it is kind of bizarre that Google, a company that literally makes a browser, built a feature around transplanting text conversations instead of leveraging the richest context source users already have.

Simon Willison wrote something last year about how the value of AI assistants correlates directly with how much of your actual working environment they can perceive, and I think that framing is exactly right. A chatbot with your imported history knows what you asked about in February. A browser agent that sees your tabs knows what you are doing right now, which is a fundamentally different kind of useful.

Data portability is solving last decade’s problem

The push for AI chat portability borrows its logic from the fight for social media data portability, which was itself about preventing monopoly lock-in. And that fight made sense when the data was the product — your photos on Instagram, your connections on LinkedIn, your posts on Twitter. Losing access to that data meant losing the thing itself.

But with AI chatbots, the data is not the product. The model’s ability to perceive and act on your current situation is the product, or at least it should be. Importing a year of ChatGPT conversations into Gemini does not make Gemini better at helping you navigate Google Cloud Console or draft a reply to the email sitting in your inbox right now. It just gives Gemini a pile of historical text that decays in relevance every single day.

Dassi takes the opposite approach. Instead of trying to reconstruct context from old conversations, it runs as a Chrome extension that reads the page you have open and works within your existing browser sessions. No importing, no transferring, no reconstructing what you meant from fragments of past chats. The context is just there because you are already logged in and already looking at the thing you need help with.

Why Google built this anyway

Probably because it is easy to build, easy to market, and it maps onto a competitive narrative that users understand. “Switching is painless” is a better sales pitch than “our model is 3% better on MMLU.” And getting people to move their data into your platform creates exactly the kind of switching cost that makes them less likely to leave later, which is ironic for a feature that is supposedly about portability.

But the engineering effort that went into building chat import could have gone into making Gemini actually aware of what the user is doing in Chrome. Google owns the browser. They own the email client. They own the productivity suite. The fact that Gemini still operates primarily as a standalone chat window, disconnected from all of that live context, is one of those strategic failures that will look obvious in retrospect.

The file cabinet vs. the window

Old chat logs are a file cabinet. Useful for reference, occasionally, if you remember what you filed and where. Your browser is a window into everything you are actively working on — every tab a live connection to a service where you are already authenticated, already mid-task, already surrounded by the context that would take paragraphs of prompting to reconstruct in a chat window.

The AI tools that will actually change how people work are the ones that look through the window instead of rummaging through the cabinet. And right now, the industry is spending a hell of a lot of energy building nicer cabinets.