|
AI Builders Digest
Monday, June 8, 2026
|
|
Token costs are suddenly everyone's biggest AI headache. Box CEO Aaron Levie says enterprises are having urgent conversations about spending they never saw coming, while former Google AI executive Madhu Guru explains why most companies are routing models completely wrong. The bill for all that AI automation is arriving.
|
|
01
|
Enterprise AI bills are getting scary
|
|
|
Box CEO Aaron Levie reports that token costs have become "one of the hottest topics for any enterprise" he talks with. Companies are using AI at scales nobody planned for, which means the costs are hitting budgets hard. But Levie sees this as bullish: it proves AI is actually working at enterprise scale, not just sitting unused after the initial demo.
|
Why it matters: Your CFO is about to start asking why the AI budget tripled this quarter. Smart companies will start optimizing for efficiency before the bills force their hand.
|
|
Source →
|
|
02
|
Why most companies waste money on the wrong AI models
|
|
|
Madhu Guru, who worked on Google's Gemini team, laid out the three phases of enterprise AI adoption. Phase 1: Everyone defaulted to GPT for everything. Phase 2: Teams over-corrected and tried to micro-optimize every task. Phase 3: Smart routing that matches tasks to the right model based on actual performance, not cost alone. Most companies are still stuck in phase 2, burning money on models that are either overkill or underpowered for their specific needs.
|
Why it matters: If you're using GPT-4 to categorize support tickets and Claude for creative writing, you're probably doing it backwards. The companies that figure out intelligent model routing first will have a massive cost advantage.
|
|
Source →
|
|
03
|
AI researchers stopped publishing because venture capital pays better
|
|
|
AI developer Swyx shared a theory about why cutting-edge AI research has largely moved behind closed doors: researchers realized they could walk out of labs and get $100 million for their knowledge instead of fighting with marketing departments to publish papers. California's ban on non-compete agreements has accelerated this trend, making it easier for researchers to monetize their expertise directly.
|
Why it matters: The days of learning about breakthrough AI techniques from research papers are largely over. If you want to know what's actually working, you'll need to watch product releases, not ArXiv.
|
|
Source →
|
|
04
|
Replit CEO faces VC backlash over political stance
|
|
|
Replit CEO Amjad Masad revealed that venture capitalists targeted him both publicly and privately after he spoke out about the Gaza conflict. While some VCs ganged up on him, others stood by him, which Masad says helped him identify "the better ones, both morally and in return profile."
|
Why it matters: Tech founders are learning that taking political stands has real business consequences. The question is whether the VCs who punish founders for their beliefs are the ones you want backing your company anyway.
|
|
Source →
|
|
05
|
Product manager discovers coding agents are "more addictive than video games"
|
|
|
Peter Yang described AI-powered coding as "agentic coding crack," comparing its addictive nature to video games in terms of engagement.
|
Why it matters: When non-technical product people are getting hooked on AI coding tools, it signals these platforms are breaking through to mainstream users, not just engineers.
|
|
Source →
|
|
Follow builders, not influencers. A daily digest of what matters in AI.
Read online ·
Archive
|