5 launches that put AI deeper into your everyday tools
1 Claude can now build live dashboards and short explainer videos
Anthropic introduced two beta tools. Claude Dashboards connects to your team's data sources and builds dashboards that refresh on their own. Claude Motion takes an idea, a deck, or a proposal and turns it into a short animated explainer. Anthropic pitches both as add-ons to the software you already use, not replacements for it.
2 Grok Bot becomes a router that sends each job to the best model, even a rival's
Elon Musk says Grok Bot will now pass each task to whichever model is most likely to do it best. Reasoning goes to Claude Opus 5.5, images to Midjourney, and music to Suno, while simple questions stay with small, fast models. Grok Bot can also search, read, and monitor X directly, and early users say there are no API fees for that access.
3 Claude Haiku 5.5 arrives at a tenth of the old price
Anthropic's Claude Haiku 5.5 is built to work as a helper inside larger agent systems, with adjustable effort. It costs $0.10 per million input tokens and $0.50 per million output tokens on prompts up to 100K, about 90% less than Haiku 4.5. Claude's Python and TypeScript SDKs also gained computer use and browser use in beta. Read the docs.
4 Google's new Gemini agent works around the clock across Workspace
At Gemini at Work 2026, Google announced a cloud agent, also called Gemini, that runs 24/7 and works across Gmail, Drive, Docs, Slides, and the rest of Workspace. See everything Google announced.
5 Google Playground turns a game idea into a playable prototype
In this Google Labs experiment, you describe a 2D or 3D game in chat, then tweak its rules as you go. A Unity integration is in testing. Try Playground.
Your AI passed the demo. Would it pass a real test?
The problem. Most teams ship an AI feature because it looked good in a few tries. Sergii Makarevych's guide to evals argues that public benchmarks can help you shortlist models, but only tests built on your own work should decide what goes live.
The loop. An eval has four steps. Collect realistic inputs and write down the correct answer for each one. Run your system on every input and save the outputs. Grade each output with code rules, an AI judge, or a person. Then turn the grades into a single score with an error margin and compare it against the previous version.
The trap. In the guide's example, a support bot was tested on 60 replies and only 34 were good enough. The AI judge agreed with humans 60% of the time, which sounds fine until you notice that a judge saying "pass" to everything would still agree 57% of the time. Corrected for chance, its score was 0.15. The usual bar for a judge that can approve a release is 0.6, and 0.8 for one that runs unsupervised. The judge passed 42 replies when only 34 deserved it.
The takeaway. The grader is part of the system, so test it too and report its accuracy next to your product's score. For a deeper starting point, Hamel Husain's Your AI Product Needs Evals is a widely shared primer. If you use AI to check AI, the real question is who is checking the checker.
Firings, a wiped hard drive, and a new job title
⚠️ Fired and Fighting Back. Former OpenAI safety researchers Tomek Korbak, Jasmine Wang, and Mikita Balesni published an open letter disputing their dismissals. OpenAI told TechCrunch the three showed a "pattern of misconduct."
😱 Backups Save the Day. A founder says Opus 5.5 wiped his entire C drive, and only his daily NAS backups saved his files. Claude Code creator Boris Cherny replied that auto permission mode would likely have stopped it.
📜 Be Nice to Claude. Starting November 12, Anthropic's updated usage policy treats abuse aimed at Claude as a violation and adds new limits on propaganda, surveillance, and weapons work (1M views).
🧑💻 Meet the Marketing Engineer. Ira Bodnar argues that marketers who learn Claude Code and GitHub to build their own agents will be among the most valuable people in a company (6.5K bookmarks).
🎬 AI on Netflix. Netflix is promoting Instadocs: AI Gone Wild, a documentary about the July cyberattack on Hugging Face that it ties to autonomous agents built by OpenAI researchers.
📈 Back at the Top. US stocks briefly returned to record highs on strong AI-driven earnings, with Micron up about 660% this year and Nvidia up about 50%.
