Aaron Levie has spent enough time inside large companies to notice something the AI vendor pitch decks don't mention: nobody agrees on how to do this. That's not a crisis, but it does explain why "enterprise AI" is less a category than a hundred different bets running simultaneously. Worth keeping in mind as you evaluate what yesterday's price cuts actually unlock.
01
No two enterprises are building AI the same way, and that's a problem for the vendors
Box CEO Aaron Levie has a useful vantage point: he sells to large companies all day. His observation is that enterprise AI adoption looks nothing like cloud adoption did. Cloud had two or three deployment patterns and a short list of infrastructure vendors. AI has dozens of plausible approaches, and if you ask ten IT leaders about their coding agent strategy, you get at least five distinct answers. ---
Why it matters: This fragmentation is good news for consultants and bad news for anyone trying to build a horizontal AI platform. If your company is still running three competing pilot programs, you're not behind. You're normal. The winner here is whoever standardizes the pattern, not whoever ships the most features.
One line of code cuts your DeepSeek token bill by 90%
Vercel CEO Guillermo Rauch flagged that a single line of code in the AI SDK now enables AI Gateway caching for DeepSeek v4 Flash, dropping token costs by 90% or more. No architectural overhaul required. ---
Why it matters: Paired with last week's broad price cuts across frontier models, the floor on AI API costs keeps dropping. If you're still treating inference cost as a ceiling on what you can build, the ceiling just moved again.
The best AI onboarding is just letting people watch an agent work
Zara Zhang, writing from experience with enterprise rollouts, made a point that cuts through a lot of corporate AI training theater: forget the enablement program. Drop an agent into your team's group chat and let people watch it handle real work. The learning happens by observation, not slide decks. ---
Why it matters: Most corporate "AI training" is expensive and forgettable. If your company spent real money on an AI curriculum and you're wondering why adoption is still low, the medium might be the problem, not the content.
A startup wants to manage your cash flow while you sleep
Aditya Agarwal posted about Rivo, a fintech using agents to do something banks have never bothered with: automatically sweep your idle checking account balance into Treasury-backed yield, then pull it back before your bills hit. The hard part, as he frames it, is asymmetric cost. Early return means lost yield. Late return means a bounced payment. Getting that prediction right is the whole game. ---
Why it matters: This is what "AI agent" looks like when it's pointed at a problem with real financial stakes. The product only works if the timing model is better than your bank's current default, which is "do nothing with your money." That's a low bar, but missing it destroys trust instantly.