Claude Dashboards and Google's Gemini agent want to do the office work for you

Claude Dashboards and Claude Motion arrive, and Docs, Slides and Design go free for everyone

If you have ever exported a spreadsheet just to build a chart for Monday's meeting, Claude can now do that job and keep it current. Claude Dashboards connects to data platforms such as Snowflake, BigQuery, Databricks, Amazon Redshift and ClickHouse, or to CRM tools like Salesforce, writes the query from a plain-English question and builds a dashboard that refreshes as the data changes. Click any number and you see the query behind it, and every chart shows when it was last refreshed, which is exactly the audit trail a finance or ops team needs before trusting an AI-built report.

Claude Motion turns a prompt into a short animated explainer, such as an animated chart, a report summary or a product walkthrough. It is built from code rather than a video model, so there is no generated footage and no AI people; you edit the result, download an MP4 or open it in Adobe, Descript, HeyGen, Luma AI or Runway. Dashboards is in beta on paid plans and Motion on Team and Enterprise.

The bigger change for most readers is the quieter one: Docs, Slides and Design are out of beta and available on every Claude plan, including Free. Standalone Claude Design is being folded into ordinary Claude chats, and the old claude.ai/design site closes on 14 December, so move any design systems across before then. It caps a busy week for Anthropic after Claude Haiku 5.5 and the new API credits on Wednesday.

Anthropic's Cyber Mission brings free AI security scans to open-source projects

If you maintain an open-source project, Anthropic will now scan it for security holes for free with its most capable models, and each report comes with a proof of concept and a suggested fix. The new OSS Scanner is opt-in and reports go out without human review, so Anthropic is candid that some will be wrong, though it expects more than 90 per cent to be genuine. Maintainers can also apply for free Claude Max through Claude for Open Source.

It sits inside a wider Anthropic Cyber Mission that also launches a Critical Infrastructure Defense Program, giving security firms such as CrowdStrike, Palo Alto Networks, Dragos and Rockwell Automation frontier Claude models and on-site engineers to help protect power grids and water systems. Project Glasswing has now been folded into the expanded Cyber Verification Program covered earlier this week. For everyone else, the practical upshot is safer open-source code underneath the apps you use every day.

$150 million puts Claude in the hands of US federal scientists

If you work in publicly funded research, frontier AI is fast becoming standard lab kit. Anthropic is committing $150 million over three years to the US Genesis Mission, providing Claude, Claude Code and API credits to several hundred research projects across more than 15 agencies, including NASA, the NIH and the National Science Foundation, with fusion energy and quantum computing named as early priorities. NVIDIA announced a $1 billion science commitment on the same day, so expect plenty more AI-assisted science stories in the months ahead.

Google unveils Gemini agent, a single "universal agent for work" that keeps going for days

If your company runs on Google Workspace, the way you use Gemini is about to shift from asking questions to handing over jobs. Announced at Gemini at Work 2026, Gemini agent takes an objective rather than step-by-step instructions, spins up sub-agents and keeps working in the cloud for hours or days after you close your laptop. You call it with @Gemini in Gmail, Docs, Sheets, Slides and Chat, and teams can create "coworker agents" with their own email address, calendar and Drive storage that only see what is shared with them.

Two details stand out. It routes each task to the best model, currently drawing on both Gemini and Anthropic's Claude, which says a lot about how multi-model enterprise AI has become. And admins can set hard per-project spending caps that pause an agent when the budget runs out, answering the most common worry about agents that run unattended. It is in private preview, with Google promising wide availability soon on select Workspace Business and Enterprise plans; Google also named its Argon, Flash, Omni and Gemma model families as the engines underneath.

Ultrafast speed tier rolls out for GPT-6.1 Sol in the API, Codex and ChatGPT Work

If you code with Codex or the API and the wait between prompt and answer is what slows you down, this is the upgrade to try. OpenAI's developer account says Ultrafast is now rolling out for GPT-6.1 Sol in the API, Codex and ChatGPT Work, promising up to eight times faster generation than standard Sol. GPT-6.1 Sol launched at DevDay as near-Astra quality for about a fifth of the price, with Ultrafast billed as coming soon, so pairing the two makes Sol a strong default for interactive coding sessions.

OpenRouter listed Sol Ultrafast about an hour later, but at the time of writing there was no OpenAI blog post or press coverage, and the announcement did not state pricing for the faster tier. Check your usage dashboard before switching long-running jobs over.

Saluki 27B squeezes a Qwen model into under 8GB under an Apache 2.0 licence

If you want a capable model running on your own laptop, this is one of the more interesting open releases of the week. Underdog has released Saluki 27B, a compressed version of Alibaba's Qwen 3.8 27B that it says is almost seven times smaller than the full-precision original while keeping 96 per cent of its benchmark performance. The company also claims it beats the original at tool calling, which matters if you want a local model to drive agents rather than just chat. It is available under the permissive Apache 2.0 licence, so commercial use is fine; the benchmark figures are Underdog's own and have not been independently tested.

Unreal Engine gets an official MCP server so AI agents can work inside the editor

If you build games or 3D scenes in Unreal, you can now let tools like Claude Code or Cursor work directly in your project rather than writing code for you to paste. Epic's Unreal MCP plugin runs a Model Context Protocol server inside the Unreal Editor, exposing actions such as spawning actors, setting up lighting, creating materials and running automation tests as tools an agent can call, and studios can add their own. It is marked experimental in Unreal Engine 5.8, with Epic warning that APIs may change, so treat it as one to explore rather than ship with today.