Edition #16. Tuesday was Google I/O and it turned into a Google week. Gemini 3.5 Flash is now the default in AI Mode, which hit one billion monthly users within a year of launch. The search box got its first real redesign since Google existed and is going multimodal. And Google launched Antigravity, a full agent stack they built themselves in twelve hours using 93 parallel agents. Outside Google, Andrej Karpathy moved to Anthropic, a talent move of a size we hadn't seen this year. Let's dive in!
Google's search box gets its first real redesign in 25 years, and AI Mode hits one billion monthly users
The biggest I/O announcement was about Search itself. Gemini 3.5 Flash is now the default in AI Mode, available in 98 languages and nearly 200 countries, no subscription needed. The search box got its first real redesign since Google existed: it goes multimodal, so text, images, files, video and even Chrome tabs work as input. Follow-up questions from AI Overviews now stay in the same flow without losing context.
The numbers Google attached are striking. AI Mode hit one billion monthly users within a year and queries double every quarter. Last quarter was the highest query volume ever on Google. In other words: the people who said AI Search would mean the end of Google have been wrong so far. Google is rebuilding under the hood, remaining the default, and even pushing volume up.
What this means for you as a business owner: if you're still thinking primarily about blue-link SEO, you're optimizing for a product that's disappearing. AI Overviews and AI Mode answer the question themselves and use websites as a source. The fight is now about being cited by the AI, not ranking at position one. That changes your whole content approach and is exactly where many companies need to revisit their AI strategy.
Antigravity: Google built a full agent stack, and built it in twelve hours with its own agents
On the same I/O, Google launched Antigravity. A desktop app, a CLI, an SDK and managed agents directly in the Gemini API. This isn't a single feature anymore but a platform to orchestrate agents, with parallel sub-agents, sandboxes and custom skills. Under the hood sits Gemini 3.5 Flash, which according to independent benchmarker Artificial Analysis runs four times faster than comparable frontier models and scores well on coding benchmarks like Terminal-Bench.
The most interesting part was how Google backed up the announcement. Antigravity itself was built in twelve hours using 93 parallel agents, 15,000 requests and 2.6 billion tokens, for less than a thousand dollars in credits. That story may be polished, but it does show what agentic development looks like a year from now. No more developer plus copilot, but an orchestrator running dozens of agents in parallel on sub-problems.
The pattern we saw earlier this year with Claude Code and OpenAI's Codex keeps playing out. The human shifts from operator to orchestrator. The question is no longer whether you put AI in your workflow, but whether your workflow is set up to manage multiple agents at once. For teams setting up their first serious agent implementation, this is the moment to design for that scale up front.
Andrej Karpathy puts his own lab on hold and joins Anthropic
Outside Google, another notable move played out. Andrej Karpathy, one of the founders of OpenAI and for years AI head at Tesla, announced he is putting his education lab Eureka Labs on hold and joining Anthropic for frontier research. That is a name on the level of Ilya Sutskever or Mira Murati. Someone who literally wrote the practical templates a large part of modern LLM training rests on.
For Anthropic this is a statement. The lab is now positioning itself explicitly as the US alternative to OpenAI and apparently, after last week's policy piece on the AI race toward 2028, has the conviction that even the best-known names in the industry are reachable. That says something about how talent is distributed right now. The pressure no longer sits only on compute and data, but on who has the people who can design the next generation of models.
Want to work smarter with AI?
Want to work smarter with AI, inside your business or just for yourself?
At AI Agent B.V. we help businesses and individuals with AI advisory, AI implementation and practical training. Not just stories about what AI might one day do, but concretely helping you use AI smarter today in your work, processes and daily operations. From strategy and tooling choices to training, implementation and custom AI agents.
Interested? Book a free half hour.
