Jason Says
Jalapeño is the real story today—OpenAI making its own inference chip isn't a tech flex, it's a direct shot at Nvidia's margins and a signal that API pricing is about to get rewritten; every SaaS business built on inference cost assumptions needs to update their models.
OpenAI and Broadcom unveiled Jalapeño, a custom inference chip purpose-built for LLM workloads. This marks OpenAI's first public foray into custom silicon, aiming to cut inference costs and reduce reliance on Nvidia—a significant vertical integration move that could reshape the AI compute landscape.
YC President Garry Tan open-sourced his personal Claude Code workflow—23 opinionated tools simulating CEO, designer, eng manager, release manager, doc engineer, and QA roles. A rare real-world, high-credibility reference for solo developers trying to ship like a team of twenty.
A developer built Drawbar, a linear workflow plugin for Claude Code that merges an accumulating knowledge base with Linear integration. Designed for slower, more deliberate coding sessions where you don't want your agent making sweeping decisions—a useful alternative to the high-throughput Beads approach.
SignalFire data reveals engineering roles have actually grown as a share of new hires, contradicting the dominant AI-kills-coders narrative. The more likely story: AI amplifies engineer output so much that companies want more of them, not fewer—a data point every developer should bookmark.
Two more top Google DeepMind researchers—Jonas Adler and Alexander Pritzel—are leaving for Anthropic, continuing a notable talent exodus that already included Noam Shazeer and John Jumper. The accelerating brain drain suggests Anthropic is winning the talent war against Google in frontier AI research.
Companies are scrambling to implement token rationing after employees burned through AI budgets on trivial tasks. Following Uber and Meta's budget blow-ups, this signals a new enterprise AI governance era—shifting from 'encourage adoption' to 'fine-grained usage control,' with implications for per-seat vs. usage-based SaaS pricing.
AGORA introduces a benchmark specifically testing LLM agents on messy, real-world workplace archives—sparse evidence, inconsistent terminology, multi-document reasoning. For developers building enterprise RAG or agent products, this is the most realistic stress-test available: if your agent fails AGORA, it will fail in production.
InSight enables Vision-Language-Action robot models to autonomously acquire new manipulation skills beyond their training data by decomposing actions into steerable primitives. Key implication for robotics developers: robots can now self-expand capabilities post-deployment without costly new human demonstrations—a meaningful step toward self-improving robotic agents.
Investors: Innovius Capital
AI 驱动医疗记录整理与患者数据结构化平台，医疗数据混乱是行业痛点，此轮资金将加速其 AI 处理管线的规模化落地。
Subscribe for daily AI updates + free playbook
📘 Subscribe Free