AI engineering, November 13, 2026

Top News

NVIDIA has named CFO Colette Kress interim CEO while the board searches for a permanent replacement for Jensen Huang, who resigned effective immediately. The board expects operations, customer commitments, and the product roadmap to continue without interruption after Jensen Huang’s immediate departure.

Here is what changed:

• Huang resigned as president, CEO, and a board member, with all three changes effective immediately.
• CFO Colette Kress became interim CEO and will continue to serve as CFO while holding the top job.
• Harvey C. Jones was appointed the board’s first-ever chair as part of the leadership changes.

One catch: Kress’s appointment is interim. The board is searching for a permanent CEO to lead NVIDIA and will consider both internal and external candidates.

Sources: nvidianews.nvidia.com

Top News · 495 HN points

Gemini 4 Hydrogen is Google’s new model for everyday coding and agent workflows. Google reports 2.5x faster task completion and more than triple the Terminal-Bench 4.1 results of Gemini 3.8 Flash at the default high setting.

Here’s what changed:

• Hydrogen surpasses Gemini 4 Argon on the Artificial Analysis Intelligence Index at default high thinking, according to Google.
• Antigravity’s Orchestrator switches between Carbon and Hydrogen, delivering near-Carbon Terminal-Bench 4.0 success with 24% less quota and 19% faster completion on average.
• Pricing is flat across the 1-million-token context window at $1.50 per million input tokens and $7.50 per million output tokens.
• Four reasoning settings, low, medium, high, and xhigh, let you trade quality against latency and cost.

One catch: Google positions Carbon and Argon, not Hydrogen, for deeper, long-horizon reasoning.

Try it: Use the Gemini API in AI Studio or Android Studio. Google Antigravity access requires a paid Google AI Pro or Ultra subscription. Also available via Vercel, releasebot.io, LiteLLM, Google AI Developers Forum

Sources: blog.google · blog.google

Top News

If your agents need to run inside your security perimeter, Anthropic now offers a self-hosted Claude Managed Agents private beta. It moves the full service into your cloud or on-premises environment, not just the sandbox.

Here’s what you can control:

• You operate the server, database, and agent workers while keeping the hosted service’s core APIs.
• Model requests can go to Anthropic, named cloud providers, or your own gateway.
• The harness calls Claude separately from the code sandbox. The agent loop, session history, and secrets stay outside that sandbox.
• Credentials can stay in your secrets manager, with a proxy adding them to supported requests. Host allowlists block other network requests.
• Session history stays in your database. Records of agent steps, tool calls, and results can be exported over OpenTelemetry for monitoring.

One catch: Access is a private beta. Contact your Anthropic account team to request it.

Sources: claude.com

1. vLLM serves DeepSeek-V4.1-Flash on NVIDIA Vera Rubin NVL72 at 58x GB300 NVL72 throughput · Reported · vllm.ai
2. SambaNova runs GLM-5.3 on SambaCloud at over 200 tokens per second
3. AWS releases agent-built AWS Emulator for testing 200+ services locally
4. Qwen Code runs open-source coding agents with runtime switching between model providers · 38 Stars today
5. MAPLE runs DeepSeek-Coder-V2-Lite at 65.6–76.3 tokens/s on NVIDIA RTX 4050 6GB laptop GPUs
6. Duo v1 world model processes audio and video while responding in real time · 7 Upvotes