Enterprise IT leaders are done experimenting with ChatGPT. Box CEO Aaron Levie just spent a week with dozens of them, and the message is clear: it's time to move from AI chat toys to agents that actually do the work.
01
Box CEO Aaron Levie: enterprises shifting from AI chat to real automation
After meeting with IT and AI leaders across banking, media, retail, healthcare, and other industries, Levie reports a major shift happening. Companies are moving beyond the "let a thousand flowers bloom" approach to AI adoption and focusing on targeted automation efforts. The transition is from simple chat tools to agents that use tools, process data, and execute real enterprise work.
Why it matters: Your company's casual ChatGPT usage is about to get serious. When enterprises stop treating AI as a productivity perk and start building it into core workflows, the budgets and expectations change completely.
Y Combinator's Garry Tan refines his agent architecture philosophy
Tan shared his latest thinking on building AI agents: put "smart fuzzy operations" into markdown skills files, keep "must-be-perfect deterministic operations" in code, and keep the framework connecting them thin. It's his clearest articulation yet of how to structure agent systems that don't break when one component fails.
Why it matters: Most companies building agents are making them too dependent on proprietary platforms. If your agent's memory dies when the framework dies, you built it wrong.
Claude Opus users report widespread quality degradation
Product leader Peter Yang noted that his entire feed and the Claude subreddit are filled with complaints that Anthropic's flagship model has been "nerfed." The reports are widespread enough that Yang publicly questioned why Anthropic would deliberately reduce their own model's performance.
Why it matters: If Anthropic's best model really got worse, it's either a technical problem they haven't acknowledged or a cost-cutting measure they haven't explained. Either way, the users paying $20/month for Opus deserve an answer.
Weil endorsed the view that human and computer intelligence aren't different points on a single line, but exist in a high-dimensional space that we should explore more broadly.