# Morning AI TL;DR — 2026-05-13

- **Google DeepMind demos an AI-native cursor.** Its May 12 “AI pointer” work turns the mouse pointer into a context-aware interaction layer for Chrome/Googlebook-style workflows, hinting at less chat-box-centric agent UX. [Source](https://deepmind.google/blog/ai-pointer/)
- **OpenAI + NVIDIA detail GPT-5.5 Codex in production.** OpenAI says Codex built on GPT-5.5 is being used by NVIDIA engineers for longer autonomous engineering workflows on GB200/GB300 infrastructure, a useful signal for where coding agents are moving next. [OpenAI](https://openai.com/index/nvidia/) / [NVIDIA](https://blogs.nvidia.com/blog/openai-codex-gpt-5-5-ai-agents/)
- **JetBrains ships TeamCity 2026.1 with MCP support.** The release adds a TeamCity CLI and Model Context Protocol integration so AI coding tools can inspect CI state and help diagnose/fix build failures directly. [Source](https://blog.jetbrains.com/teamcity/2026/05/teamcity-20261/)
- **OpenAI moves deeper into enterprise services.** Bain announced investment in the OpenAI Deployment Company, a new venture with 19 partners aimed at embedding AI systems into critical business operations—relevant to Salesforce/enterprise builders competing on implementation depth, not just models. [Source](https://www.bain.com/about/media-center/press-releases/2026/bain-company-openai-a-new-venture-to-deploy-ai-at-enterprise-scale/)
- **Baidu releases ERNIE 5.1 with a cost-efficiency story.** Reports say the new model is distilled from ERNIE 5.0 and materially reduces pre-training cost while competing near frontier Chinese benchmarks, reinforcing the “cheaper capable models” trend. [Source](https://the-decoder.com/baidus-ernie-5-1-cuts-94-percent-of-pre-training-costs-while-competing-with-top-models/)
- **Copilot Chat security research flags agentic edit risks.** Hacktron published an RCE/bypass write-up involving VS Code Copilot Chat’s file-edit confirmation flow, a reminder that coding-agent guardrails need OS/repo-level containment, not just UI approvals. [Source](https://www.hacktron.ai/blog/rce-in-vscode-copilot)
- **Anthropic spins out Petri alignment tooling.** Anthropic transferred its open-source Petri evaluation/alignment tool to Meridian Labs, strengthening independent eval infrastructure for labs, governments, and builders. [Source](https://www.anthropic.com/research/donating-open-source-petri)

**Worth watching:** Enterprise AI differentiation is shifting from “which model?” to “which workflow substrate?”—CI/CD, cursor/UI context, deployment teams, and sandbox/security layers are becoming the real moat.