Aaron Levie has been making the same argument for three days running, and each version gets sharper. Yesterday it was about workflow redesign. Today it's about something more fundamental: the companies that win the AI race won't be the ones with the best models. They'll be the ones with the best data about you.
01
The AI platform war is actually a war over who knows your business
Box CEO Aaron Levie laid out a clean thesis: agent effectiveness comes down to context, not capability. The model is almost a commodity at this point. What matters is whether the agent has domain expertise, access to the right tools, and enough understanding of your specific workflows to be useful rather than just plausible. The platforms that capture the richest context about how companies actually operate will have a structural advantage that's very hard to copy. ---
Why it matters: Your CRM, your Slack, your email, your code repos, these aren't just software subscriptions. They're context databases. Whoever owns that context owns the AI layer on top of it. Microsoft 365 users, Salesforce customers, and Box enterprise clients are already inside this battle whether they know it or not.
Vercel CEO makes the case for agents that fix themselves
Vercel CEO Guillermo Rauch posted about "agentic self-improvement": the idea that agents should be able to review their own past runs, identify where they wasted time or made errors, and rewrite their own instructions accordingly. He's using it as a pitch for Vercel's built-in agent observability, but the concept stands on its own regardless of platform. ---
Why it matters: Right now, when your AI agent does something stupid, a human has to notice, diagnose it, and fix the prompt. That's a hidden tax on every team running agents in production. If agents can close that loop themselves, the maintenance cost of running AI workflows drops dramatically and the gap between "impressive demo" and "reliable product" shrinks.
Cat Wu shared a short tip: you can use Claude Code with computer use to automate the entire Claude Tag onboarding process. Point it at the docs and it will connect your GitHub repo, data warehouse, Google Drive, and other sources on its own. ---
Why it matters: The friction of connecting data sources is usually where enterprise AI pilots stall. A setup process that runs itself removes one of the more tedious excuses for not getting started.
The prettiest productivity tools lost to a command line
Swyx, who organizes the AI Engineers community, posted a sharp observation: "tools for thought" builders spent years making beautiful visual interfaces with infinite canvases and spatial note-taking, and then got completely beaten by command-line AI tools that are ugly but actually do the thinking for you.
Why it matters: There's a product lesson here that runs counter to everything taught in design school. Users will tolerate a terrible interface if the output is good enough. The teams building polished AI wrappers should take note: if the underlying capability doesn't do real work, no amount of UI will save you.