Corporate finance teams are about to discover that AI agents eat budgets like developer laptops used to eat desk space. The difference: you can't just buy tokens in bulk and forget about them for three years.
01
Box CEO Aaron Levie warns of the coming token budget wars
Box CEO Aaron Levie highlighted a trend hitting larger enterprises: token budgeting is becoming a major operational challenge. As AI agents handle longer-running tasks that consume vastly more compute, companies are spending serious time allocating token budgets across teams, just like they currently manage spending on talent, marketing campaigns, and laptop setups.
Why it matters: Your CFO is about to start asking why the engineering team burned through 50 million tokens last month. Companies that figure out token governance early will have a real advantage over those treating AI usage like an unlimited buffet.
Investor Matt Turck questions pure consumption pricing for AI agents
Investor Matt Turck argued that AI agents might not fit purely consumption-based pricing models. While token costs matter, enterprise agents need identities, roles, authentication, budgets, and audit logs. He noted this infrastructure "sounds oddly seat-like," just not for humans.
Why it matters: SaaS companies built entire business models around per-seat pricing. If agents need similar infrastructure as employees, expect hybrid pricing models where you pay both for seats and consumption.
Developer builds YouTube copilot that watches videos with you
Developer Zara Zhang created a browser extension using OpenAI's realtime 2 API that acts as a YouTube viewing companion. The AI agent watches videos alongside users and answers questions about what was just said via voice chat. The standout feature: it can differentiate between the YouTube audio stream and the user's voice, so it doesn't confuse video content as commands.
Why it matters: This is what ambient computing actually looks like. When AI can separate your voice from background noise in real-time, every screen becomes a potential conversation partner.
Anthropic's Amanda Askell shifts focus from preventing bad AI to building good AI
Anthropic alignment researcher Amanda Askell reflected on moving beyond just averting concerning AI behaviors toward giving models "an honest and positive vision for what AI models can be and why." She called it an exciting future direction for alignment research.
Why it matters: This signals a major philosophical shift at Anthropic from "stop AI from doing bad things" to "teach AI to actively do good things." The difference could determine whether future AI feels like a helpful colleague or an overly cautious legal department.