Google's I/O conference revealed the company betting hard on remaining relevant in search by abandoning what made it dominant. The AI Mode, rolling out to US users, embeds a conversational chatbot directly into the search bar—essentially admitting that users want answers, not lists of links. The immediate consequence is stark: fewer clicks to websites, less attribution to sources, weaker revenue for publishers already drowning in competition. Google's bet is that AI Overviews and direct answers will keep users inside its ecosystem long enough to serve ads on the results themselves, but this inverts the entire logic of the ad-supported web. When 41% of traffic to some sites now flows from chatbots rather than Google's organic search, we're watching the gatekeeping function shift, not disappear. Who benefits? Only platforms able to aggregate and resell access.
This redistribution is why Search Engine Optimization for LLMs—a term barely two years old—matters more than ever. Websites now compete not for human eyes but for tokens in training datasets. Podcasts without transcripts, articles hidden behind logins, content locked in proprietary platforms: all become invisible to the systems that will drive traffic tomorrow. The arithmetic is brutal: fewer visits but potentially more engaged ones, and whoever controls the ranking gets the take rate.
Separately, OpenAI, Copilot, and Google all launched coding agents this week—autonomous systems that manage repositories, run tests, deploy code, and handle pull requests without human intervention. These aren't autocomplete or no-code builders. They're estagiários with infinite parallel capacity, able to tackle multiple features simultaneously while a developer oversees the work. It's software eating software development, and it's moving fastest in the one industry that foundation models were optimized for from day one.
Yet the most consequential announcement received minimal attention: CRISPR gene editing has moved from theory to patients. Two cases—a newborn and an adolescent—have been treated for rare genetic mutations, with promising early outcomes. A single therapeutic dose costs roughly $2 million and requires custom engineering for each patient's specific mutation. That bottleneck will eventually dissolve when AI systems can design therapies faster than clinicians can administer them. We are watching the first steps toward curing genetic disease at the source rather than managing its symptoms for life.