# Morning AI TL;DR — 2026-05-17

Fresh high-signal AI news was a bit thin over the weekend, so this includes the most relevant updates from roughly the last 24–48 hours plus a few recent ecosystem releases worth tracking.

- **xAI beta-tests Grok Build for coding agents** — Early coverage says xAI’s new Grok Build targets terminal/web coding workflows with planning, parallel agents, and MCP/ACP support, a direct shot at Claude Code, Cursor, Devin, and Copilot agents ([AIbase](https://news.aibase.com/news/28016), [Zypa](https://zypa.in/grok-build-cli/)).
- **NVIDIA releases SANA-WM, an open world model** — The 2.6B-parameter model reportedly generates minute-scale 720p video on a single GPU, useful for robotics/simulation teams watching low-cost “physical AI” tooling ([MarkTechPost](https://www.marktechpost.com/2026/05/16/nvidia-introduces-sana-wm-a-2-6b-parameter-open-source-world-model-that-generates-minute-scale-720p-video-on-a-single-gpu/)).
- **Salesforce tightens Agentforce dev lifecycle** — New Agentforce metadata and Builder lifecycle updates aim to make agent authoring/test/deploy safer and more versionable for enterprise teams ([Salesforce Developers](https://developer.salesforce.com/blogs/2026/05/new-agentforce-metadata-and-development-lifecycle)).
- **Agentforce debugging gets visual tooling** — AgentLens adds interactive tracing for Agent Script routing, variables, and reasoning loops, which should reduce “black box” debugging pain in production agent builds ([Salesforce Developers](https://developer.salesforce.com/blogs/2026/05/agentlens-debug-agentforce-with-interactive-visualizations)).
- **Salesforce Vibes leans into intent-driven tool selection** — “Abilities” are positioned to help AI coding agents choose the right Salesforce DX/MCP tools automatically, a practical step toward safer enterprise vibe coding ([Salesforce Developers](https://developer.salesforce.com/blogs/2026/05/intent-driven-tool-selection-using-abilities-in-agentforce-vibes)).
- **GitHub Copilot’s coding agent remains the benchmark to watch** — GitHub’s background coding agent can take issues, run in Actions, and open PRs, keeping pressure on standalone coding-agent vendors ([GitHub Blog](https://github.blog/news-insights/product-news/github-copilot-meet-the-new-coding-agent/)).
- **OpenAI product strategy shifts back toward Brockman** — TechCrunch reports Greg Brockman is taking charge of product strategy, notable as OpenAI pushes GPT-5.5/Codex-era productization and developer tooling ([TechCrunch](https://techcrunch.com/2026/05/16/openai-co-founder-greg-brockman-reportedly-takes-charge-of-product-strategy/)).

**Worth watching:** Enterprise agent platforms are converging on the same primitives — versioned metadata, traceable execution, protocol-based tools, and background PR/workflow agents — so the near-term edge will be governance and debugging, not just model quality.