The data shortage problem just got real. While everyone's been obsessing over token costs and model routing, the real constraint on AI progress is lurking in plain sight: we've run out of easy training data, and the hard stuff requires humans who actually know what they're doing.
01
Former Google engineer explains why we have coding agents but not lawyer agents
Madhu Guru, who worked on Gemini at Google, points out a critical misconception about AI training data. The frontier models need data for high-value tasks, but most knowledge work outside of software engineering has little documentation. It's complex, domain-specific knowledge built over years using legacy tools that don't integrate well. Software engineering, by contrast, has decades of documented code, tutorials, and Stack Overflow answers.
Why it matters: This explains why your company's AI coding assistant works great but the AI lawyer is still useless. The data bottleneck means we'll see AI agents roll out in very specific order, and legal, medical, and finance work will be last, not first.
Claude Opus tips for running AI agents for days without supervision
Boris Cherny shared five specific techniques for running Anthropic's Opus model autonomously for hours or days: use auto mode so Claude doesn't ask for permissions, use dynamic workflows to orchestrate hundreds of agents, use /goal or /loop commands to keep Claude working until completion, use Claude Code in the cloud so you can close your laptop, and make sure Claude has sufficient usage limits.
Why it matters: These aren't theoretical tips. Cherny is describing actual workflows that let you start an AI project on Monday and check the results on Wednesday. The bottleneck is shifting from "can AI do this?" to "do you trust AI to do this unsupervised?"
Box CEO Aaron Levie on why AI won't kill enterprise software
Aaron Levie argues the market misunderstood how AI would impact enterprise software companies. While AI makes building software somewhat easier, the majority of costs at enterprise software companies actually go to sales and marketing, not engineering. Most enterprise software categories are competitive on distribution and relationships, not just product quality.
Why it matters: All those "AI will replace Salesforce" predictions missed the point. Your AI startup might build better software than Salesforce, but Salesforce has 8,000 salespeople and relationships with every Fortune 500 CIO. That doesn't disappear because you have a better algorithm.
Anthropic's Codex program launches 100-day power user experiment
Anthropic's Thibault Sottiaux announced they'll select one person per day for the next 100 days who does impressive work with Codex and give them 10x usage limits for a month to see what they can accomplish with unlimited access.
Peter Yang asked Twitter to finish his AI-themed dad joke: "My wife asked me to take out the trash. I said, 'You shouldn't be prompting me anymore. You should be designing loops.'"