The models and the money kept sprinting this week. Google's Gemini crossed a billion monthly users, xAI shipped another frontier-class model, and a code review startup raised $143 million on the strength of one idea: somebody has to check all the code AI is now writing. The same week made the cost of that pace visible, as Google's DeepMind, the lab behind much of this, lost its founder-CEO and a wave of senior talent. Here is what mattered.
Google's Gemini crossed a billion monthly users
On August 11, Sundar Pichai said the Gemini app had passed one billion monthly active users, making it the fastest product to a billion in Google's history and the company's fourteenth to reach that mark. The app went from 400 million users in May 2025 to a billion in a little over a year. Google says 63 percent of users interact by voice, and the app now generates more than 150 million images a day. [TechCrunch, The Information]
Our take: a billion people now reach for an AI assistant by default, and most of them talk to it out loud. That is the real signal in this number for any business. Your customers are getting comfortable speaking to software, which lowers the bar for putting a capable voice or chat agent in front of them. You will not build your own Gemini, and you do not need to. The opening is to use tools this good as the engine and wrap them around your specific workflow, where your knowledge of the customer is the advantage.
The lab behind much of modern AI is under strain
Demis Hassabis is stepping down as CEO of Google DeepMind to become Alphabet's chief scientist and DeepMind's chairman, with CTO Koray Kavukcuoglu taking over daily operations. A detailed Fortune report this week tied the shakeup to stalled models, missed deadlines, and staff burnout, and noted that chief scientist Jeff Dean and several colleagues are leaving to start their own company, which Google plans to back. [Fortune, Axios]
Our take: even the best-funded lab in the world is finding that this pace has a human cost. That is worth holding onto the next time a vendor promises to move at AI speed. Deadlines slip and people burn out at every scale. The lesson for your own team is to buy the capability and rent the roadmap, rather than betting your business on one lab's next release landing on time. We design around what ships and works today, with room to swap in whatever comes next.
xAI shipped Grok 4.6
On August 12, xAI released Grok 4.6, about five weeks after 4.5, with a focus on long-running agent tasks and a claim of matching OpenAI's GPT-5.6 Sol on a composite benchmark. It keeps a headline price of $2 per million input tokens and $6 per million output. That rate holds only for prompts under 200,000 tokens, though; above that, reporting on xAI's pricing puts the whole request at a higher tier of $4 and $12. xAI says a larger model, Grok 4.7, is due within weeks. [xAI, Digital Applied]
Our take: the model race is measured in weeks now, not quarters, and the prices keep drifting down. The trap is the fine print. That $2 rate becomes $4 the moment your prompt crosses a size threshold, which is easy to do once you start packing context into an agent. When we build on these models, we watch the real token bill rather than the headline rate, and we shape prompts to stay in the cheap tier, because a low per-token price only helps if you control how many tokens you use.
A startup raised $143 million to check the code AI is writing
CodeRabbit, an AI code review company, raised a $143 million Series C at a $1.5 billion valuation, co-led by Atomico and Smash Capital, with Datadog and BMW i Ventures among the new investors. The company says it now runs more than two million code reviews a week for 17,000 customers, including NVIDIA and BMW, and it used the raise to launch a governance layer for software written by both people and AI agents. [PYMNTS, Yahoo Finance]
Our take: this is the whole game in one funding round. As AI writes more of the code, the scarce thing becomes trustworthy review of it, and a billion-dollar valuation is riding on that being a durable need. We think it is. You do not need CodeRabbit's scale to apply the principle: every line an agent writes still goes through review before it ships, and a person owns what merges. We build that checkpoint into the workflow from the start, because AI that writes code fast is only an asset if something trustworthy is checking it.
The one thing to remember
The models got faster and cheaper again, and the money kept betting on more of the same. The harder problem this week was human. A leading lab strained under its own pace, and the fastest-growing tools were the ones that help people keep up with what AI produces. The teams that do well with this will be the ones who pair the new capability with the boring parts: review, limits, and a clear owner for anything that ships.