AI Builders
Digest

The Takeaway: Intel under Lip-Bu Tan is strengthening its balance sheet, simplifying products, listening to customers, and capitalizing on surging CPU demand driven by agentic AI to regain leadership. Lip-Bu Tan, legendary investor, former Cadence CEO, and now Intel CEO, brings startup speed and humility to one of America's iconic semiconductor companies. He emphasizes crawling before running: fix the balance sheet with government and NVIDIA support (Jensen Huang's investment turned $5B into $25B), simplify product lines, and drive next-gen leadership products. Tan sees CPUs gaining ground against GPUs (from 1:8 to potentially 1:1 ratio) for orchestration and reinforcement learning in agents. He is pushing foundry improvements in yield, IP, and advanced packaging while collaborating with Elon Musk on TeraFab for massive capacity. Tan stresses resilient supply chains, new materials like gallium nitride and glass substrates, and full-stack solutions. 'Nine of the 10 company I invest, halfway they change their business plan because market have changed.' He remains optimistic about US manufacturing and AI-driven efficiency gains across the industry.

Box CEO Aaron Levie dives into a central AI debate: if closed stacks maintain a wide frontier lead, vertical integration works; but if open weights stay close, the majority of tokens will flow to alternative stacks controlled by others. This has major implications for regulation and control. Levie highlights how beliefs about open weights' proximity to frontier intelligence shape views on AI governance.

Anthropic Engineering: Scaling Managed Agents decouples the brain (Claude + harness) from hands (sandboxes) and session log for reliability and flexibility. New features include self-hosted sandboxes (Cloudflare, Daytona, Modal, Vercel) and MCP tunnels for private network access. This allows agents to run securely within enterprise perimeters while maintaining long-horizon capabilities. Anthropic Engineering: An update on recent Claude Code quality reports details fixes for reasoning effort defaults, thinking cache bugs, and verbosity prompts that temporarily impacted quality; all resolved with resets and improved processes.

Claude Code @anthropicai engineer Boris Cherny announces that in the next version of Claude Code, subagents will run in the background by default, allowing continued conversation with Claude while they work. Users can specify foreground execution if needed.

OpenAI's Thibault Sottiaux details Codex usage limit resets and fixes: auto-review changes, subagent work, and background suggestions were overconsuming. Improvements include better reporting, reverted changes, and monitoring. Advanced permission profiles for least-privilege sandboxes also shipped.

Vercel CEO Guillermo Rauch highlights 20x larger functions support on Vercel and teases more deployments. Demo shows streaming capabilities.

Swyx reports strong attendance at non-lab AI workshops despite competing OpenAI sessions, showing hunger for practical tools from Snyk, Atlassian, Neo4j, Arize, Akamai, and Together. Also shares humorous scaling laws reminder.

Builder Zara Zhang released an open-source Chrome extension that auto-books calendar time for your read-later list. Local, no server, links included.

Former Google PM Madhu Guru notes that strong open-weight models like GLM will drive more fine-tuning experiments, benefiting Google Cloud's managed platform, reliability, and compute stack.

Anthropic's Thariq describes his writing process: engineering work, discussions, Claude brainstorming, drafting, talks, rewriting, and early morning final polish.

Peter Yang finds plain Claude web best for writing/editing over coding agents, and shares spicy takes on restricted access and open source.

Y Combinator CEO Garry Tan urges building power and datacenters to support AI growth.

Claude Opus 4.8 and Haiku 4.5 now GA on Azure with enterprise features.