The Clippy cycle of AIs

It all started with Clippy, the paperclip with eyes that Microsoft launched in the 1990s to assist novice Windows users. It’s a shame that Clippy was extremely limited and ended up annoying anyone who saw it pop up randomly on their screen offering a helping hand. The response was always the same: no, thank you. When it was withdrawn in 2001, few people mourned its loss (and since then, Clippy has survived only as an ironic meme for the nostalgic). Almost thirty years on, the cycle seems to be repeating itself. The paperclip has grown up and been reincarnated as the artificial intelligence that companies such as OpenAI, Google, Microsoft and Apple are pushing into every one of their products. The problem is that this has generated a sort of collective performance anxiety: AI has to be everywhere – on every app, every window, every toolbar. And its presence isn’t always welcome.

The Microsoft case is the most instructive. In the second half of 2025, the company had been criticised for compromising the Windows 11 user experience in order to promote Copilot, its AI assistant. Copilot popped up everywhere: in system windows, in pre-installed apps, in Excel, in menus. Worse than Clippy, they said – and that was no compliment. Microsoft eventually backtracked last March, removing or scaling back many of these integrations. The press release announcing the U-turn stated the intention to ‘integrate AI where it is most useful’: in corporate jargon, a mea culpa. One would expect such a precedent to act as a deterrent. That hasn’t been the case, judging by what Google is doing: it has recently updated Gemini, integrating it even more deeply into its suite. Gemini is everywhere – on Gmail, Google Calendar, Google Drive, Google Docs – in the form of a star-shaped icon begging to be clicked. In itself, the model works: open the chatbot, ask a question, get an answer. It is the integration into other interfaces that is causing problems, especially following this latest update.

The point is that you don’t need the most powerful model on the market to translate a sentence or rephrase a paragraph. Yet the new Gemini has been enhanced to such an extent that any request, however trivial, triggers a lengthy thought process: context analysis, connecting to Google Drive, reading emails. This is particularly evident in Google Docs: until a few weeks ago, users on the AI Pro plan could select text and receive suggestions in real time. Now, however, you have to wait whilst Gemini ‘reasons’ and consults data from Google Drive or Gmail, even for the most basic tasks. It’s the curse of Copilot, which is ultimately the same as Clippy’s: the moment when an assistant becomes so intrusive that it negates any positive contribution it might have made.

The question is who will be next, but perhaps the ideal candidate is Apple, which this week unveiled the new Siri, enhanced by artificial intelligence. To make the project work, Apple has struck a deal with Google to integrate Gemini into Apple Intelligence, creating Siri AI. The cycle has therefore already begun: a company creates its own AI and is eager to impose it, at the risk of pushing it too far and worsening the experience for its users. There are those who think Apple is immune to mistakes of this sort. That may well be the case, but recent years have shown that infallibility is no longer quite so much the norm in Cupertino: following the initial failed launch of Apple Intelligence and the Vision Pro debacle, we can expect anything. Perhaps Siri AI will follow the same path as Copilot: it will take off, mess things up, then scale back. It’ll be somewhat inconvenient for a period. But it’ll blow over eventually.