That's a very valid concern, especially for content-heavy sites (blogs, wikis). If an LLM can summarize the answer, the user has no reason to click.

However, I suspect there's a distinction between Information and Utilities.

An LLM can summarize a StackOverflow discussion on "how to compress a PDF," but it cannot (yet) reliably perform the heavy client-side processing to actually do it securely in the browser without uploading data.

For tools and utilities, the "click" is still necessary to perform the action. My bet is that AI will act more as a dispatcher for specific tasks ("Go here to fix X") rather than just a summarizer.

But you're right — once LLMs get native, sandboxed execution environments, even tools might get absorbed.