Box CEO Aaron Levie called new ChatGPT agents "the biggest news yet in software going headless." Meanwhile, Vercel CEO Guillermo Rauch is dealing with the messy reality of what happens when AI systems get compromised. The gap between the AI future we're building and the security problems we're ignoring keeps getting wider.
01
Anthropic lands its biggest enterprise deal yet with NEC
NEC will deploy Claude to 30,000 employees across Japan and become Anthropic's first Japan-based global partner. The companies plan to co-develop AI products specifically for finance, manufacturing, and government sectors. This marks Anthropic's most aggressive push into enterprise markets outside the US.
Why it matters: Anthropic just leapfrogged OpenAI in Japan's enterprise market. While OpenAI focused on consumer ChatGPT adoption, Anthropic built the relationships that actually pay the bills. Every other AI lab is now scrambling to copy this playbook in other regions.
Box CEO Aaron Levie: ChatGPT agents will bring "headless software" to everyone
Box CEO Aaron Levie called OpenAI's new agent capabilities "probably the biggest news yet in software going headless." He demonstrated a custom sales assistant agent that uses Box as a secure knowledge source to answer questions and generate content on demand. Levie emphasized that these agents can access any tools and data you want, with complete coding and tool use capabilities built in.
Why it matters: Your company's software stack is about to get a lot simpler. Instead of training employees on 15 different apps, you'll have agents that know how to use all of them. The SaaS companies that don't build agent-friendly APIs are going to lose customers fast.
Swyx explains why GPT's new image model takes 10+ minutes to generate
AI researcher Swyx clarified that GPT-Image-2-Thinking isn't just a slow image generator. It's actually an image agent that uses search and Photoshop-like tools in a loop, searching and compositing and reviewing its own work until it gets the result right. That's why it can generate perfect QR codes, diagrams, and logos in one shot, but takes multiple tens of minutes to do it.
Why it matters: Stop comparing this to Midjourney speed. You're paying for an AI designer that iterates on its work, not an AI printer that spits out random images. The 10-minute wait time suddenly makes sense when you think of it as hiring a freelancer who actually checks their work.
Vercel CEO updates on security breach investigation
Vercel CEO Guillermo Rauch provided details on the ongoing security investigation following the Context.ai compromise. The team analyzed nearly a petabyte of logs across Vercel's entire network and discovered the threat actor was active beyond the initial startup compromise, with threat intelligence pointing to broader distributed activities.
Why it matters: The AI supply chain is more interconnected than anyone realized. When one AI startup gets hacked, the blast radius includes every platform they touched. If you're building on multiple AI services, you need to assume they're all potentially compromised.
Replit CEO Amjad Masad shared that Replit's Security Agent is now automatically reviewing apps, claiming it's "making the internet a better place one app review at a time." No details on what the agent actually catches or how it integrates with Replit's development workflow.