The most famous AI researcher just picked sides in the biggest battle in AI. And while everyone's watching that move, enterprise customers are quietly panicking about something much more practical.
01
Andrej Karpathy joins Anthropic
The former Tesla AI director and OpenAI co-founder announced he's joining Anthropic to work on R&D at "the frontier of LLMs." Karpathy called the next few years "especially formative" and said he plans to resume his education work later. This comes after he spent time as an independent AI educator following his departure from Tesla.
Why it matters: Karpathy is one of the few people who helped build the original transformer architecture that powers every major AI model. His choice to join Anthropic over OpenAI, Google, or staying independent signals where top talent thinks the most interesting AI research will happen next.
Google ships Gemini Spark, a 24/7 personal AI agent
Google product manager Josh Woodward announced Gemini Spark, described as a proactive AI agent that manages tasks and helps navigate digital life "all under your direction." It starts rolling out to trusted testers this week and hits beta for US Google AI Ultra subscribers next week.
Why it matters: This is Google's answer to the "AI that does things for you" race that every tech giant is running. The phrase "proactive" suggests it will take actions without being asked, which is either the future of personal computing or a privacy nightmare waiting to happen.
Venture capitalist Matt Turck highlighted Google's latest model performance, showing major improvements in coding tasks: 76.2% on Terminal-Bench for agent-based coding and 56.5% on real-world task benchmark Toolathon. The model also scored leading results across SWE-Bench and other developer benchmarks.
Why it matters: Google was considered behind in AI coding tools six months ago. These numbers suggest they've not only caught up but are setting the pace for AI that can actually write and debug code autonomously.
Enterprise CIOs are freaking out about AI token costs
Box CEO Aaron Levie reported that token costs dominated a recent dinner with Fortune 500 CIOs, calling it "the most heated topic." Companies are scrambling with strategies like routing workloads to different models based on user type, setting spend caps by team, and requiring teams to justify AI usage by specific use cases. Levie noted that "basically no one feels like they have the right solution."
Why it matters: Your company's AI budget is about to become as political as the travel budget used to be. Expect "token management" to become a real job title and a lot of awkward conversations about why the marketing team burned through $50,000 of Claude credits in a week.
Product wisdom goes viral: 90-day roadmaps are the new normal
Former Meta PM Peter Yang shared quotes from an unnamed product leader about modern development cycles: "We only have a 90 day roadmap and maybe if we are lucky it's 120 days. I don't know if we'll ever go back to 1 year roadmaps. I haven't been working on one of those in 5 years." The thread emphasized rapid iteration and building to learn.
Why it matters: If you're still planning AI features on annual roadmaps, you're planning for a world that no longer exists. The companies winning in AI have shifted to quarterly sprints because the technology changes faster than traditional product cycles can handle.