Token costs are now eating enterprise AI budgets faster than anyone expected. And the scramble to optimize spending is creating a whole new layer of complexity that most companies aren't ready for.
01
Box CEO Aaron Levie: Token costs become the new cloud bill shock
Aaron Levie reports that token expenses are now the hottest topic in every enterprise AI conversation he's having. Companies are discovering their AI usage has scaled far beyond what anyone anticipated, creating unexpectedly large bills. But Levie sees this as bullish for AI adoption overall, since it means real usage at real scale. The solution, he argues, will be "model routing" — automatically sending different tasks to different AI models based on cost and quality tradeoffs.
Why it matters: Your CFO is about to start asking pointed questions about AI spending. The companies that figure out smart model routing now will have a massive cost advantage over those still defaulting everything to GPT-4.
Former Google engineer maps the three phases of enterprise AI adoption
Madhu Guru, who worked on Gemini at Google, shared how he watched enterprises evolve through predictable phases: first, defaulting to GPT for everything because it was trendy; then over-correcting by obsessing over tiny cost differences; finally reaching a mature approach of benchmarking models against specific tasks. The hard part isn't the technology — it's mapping each workflow to the right model and dialing in the quality versus cost balance.
Why it matters: If your company is still in Phase 1 (using the same model for everything), you're burning money. If you're in Phase 2 (optimizing costs without measuring quality), you're probably delivering worse results to save pennies.
AI developer Swyx: Research stopped when $100M became easier than publishing
Swyx argues that AI research publication essentially died when researchers realized they could walk out the door with their knowledge and raise nine-figure funding rounds instead of fighting with marketing departments over paper releases. He points to California's ban on non-compete agreements as having more impact on AI knowledge sharing than GitHub, arXiv, and Hugging Face combined.
Why it matters: The best AI research is now happening behind closed doors at startups, not in public papers. If you're waiting for the next breakthrough to be published, you're already 18 months behind.
Replit CEO Amjad Masad on standing up to investor pressure
Amjad Masad reflected on facing backlash from some VCs for his public stance on Gaza, noting that the investors who supported him turned out to be both morally superior and better business partners. His advice: let people self-select out of your life by standing for your beliefs.
Why it matters: The AI industry's rapid growth is creating new power dynamics between founders and investors. The CEOs willing to take principled stands may end up with better backing in the long run.