AI Leadership Weekly · Issue #66 · Tuesday 13 January 2026 · 08:00 GMT

Good morning.

The inbox, desktop folder and checkout are all in the news. Anthropic introduced Cowork, Google announced Gemini-powered Gmail changes, and a new commerce protocol set out how agents might work with retailers. These developments sit close to everyday activity, rather than inside a specialist AI team. Their initial access conditions matter, especially for UK organisations reading US-led launch announcements.

Claude introduces Cowork. Anthropic launched Cowork as a research preview for Max subscribers on macOS, allowing work with files in an authorised folder.

Gmail adds Gemini features. Google announced new Gemini-powered Gmail features, with initial US-English access and different availability across free, paid and testing experiences.

Google proposes a commerce protocol. Google introduced the Universal Commerce Protocol for agent-led shopping, alongside checkout plans and new tools for retailers.

https://support.claude.com/en/articles/12138966-release-notes

Anthropic · 12 January 2026

What happened. Anthropic’s 12 January release notes introduced Cowork as a research preview for Max subscribers using Claude Desktop on macOS. The notes describe an isolated local environment with access to local files and connected tools. This initial release should not be confused with later plans, operating systems or enterprise capabilities.

Why it matters. Our take: File access changes the character of a pilot. The assistant is no longer limited to proposing text for someone to copy. Give a trial a deliberately narrow working area and decide what should happen when a request is ambiguous. A familiar desktop interface is not evidence that your organisation has approved the workflow.

What to do. Use copies of non-sensitive files in a dedicated test folder. Define the expected outputs and preserve the originals. Review every changed file before using it. Ask the team to report uncertainty or unexpected actions, not just examples where the assistant successfully completed a task.

https://blog.google/products-and-platforms/products/gmail/gmail-is-entering-the-gemini-era/

Google · 8 January 2026

What happened. Google announced new Gemini-powered Gmail features on 8 January, including conversation summaries, writing assistance and paid natural-language inbox questions. It described an initial US English rollout and an AI Inbox being tested with selected users. Different features had different access conditions; the announcement was not blanket availability for every business account.

Why it matters. Our take: Treat a summary as a route into the evidence. A commercial promise may depend on the latest reply, a qualification in an attachment or a change of scope. In a team process, make it easy to open the original message. Do not let a polished summary silently become the authoritative record of a commitment.

What to do. Test a small set of completed email threads with known outcomes. Include a changed deadline and a withdrawn proposal. Ask users to check the summary against the original exchange and record what it misses. Keep sending messages and confirming commitments under the existing approval process.

https://blog.google/products/ads-commerce/agentic-commerce-ai-tools-protocol-retailers-platforms/

Google · 11 January 2026

What happened. On 11 January, Google announced the Universal Commerce Protocol with retail and platform partners. It described upcoming checkout for eligible US product listings in AI Mode and Gemini, while retailers would remain the seller of record. The announcement also introduced related retailer tools; it did not make agent-led checkout universally available.

Why it matters. Our take: The practical preparation is product and service clarity. An assistant needs reliable information about what is being sold, what it costs and what happens after purchase. Keep the experience connected to your existing order and support processes. A new discovery surface should not create a parallel version of the customer promise.

What to do. Review a small set of important products or services for inconsistent descriptions, availability and fulfilment information. Name the owner of each data field. Map how an order or customer query would reach the existing operation before investing in an additional conversational sales interface.

22 December · Atlas receives security hardening. OpenAI described further defences against prompt injection in Atlas, treating malicious webpage instructions as an ongoing security problem. Source

18 December · GPT-5.2-Codex reaches paid users. OpenAI introduced GPT-5.2-Codex for eligible paid ChatGPT users; broader API availability was still forthcoming in the launch announcement. Source

17 December · Coursera and Udemy plan to combine. Coursera and Udemy announced a proposed combination. Completion remained subject to the transaction’s conditions and approvals. Source

17 December · Gemini 3 Flash arrives. Google released Gemini 3 Flash, broadening access to its latest model generation with an emphasis on speed and cost. Source

More dated updates from the preceding 30 days.

11 January · Claude expands its healthcare tools. Anthropic announced healthcare and life-sciences connectors and workflow tools. The launch included specialist administrative applications, not evidence of autonomous clinical judgement. Source

7 January · OpenAI introduces ChatGPT Health. OpenAI announced a dedicated health-and-wellness experience with staged access. It was designed to support, not replace, care from qualified professionals. Source

5 January · NVIDIA sets out Rubin. NVIDIA announced the Rubin platform and its deployment roadmap. Future infrastructure claims did not establish immediate savings for customers. Source

24 December · Groq licenses technology to NVIDIA. Groq announced a non-exclusive inference-technology licensing agreement with NVIDIA. It was not described as a sale of the entire company. Source

Minutes 0–4 · Choose a familiar process. Pick a task involving files, messages or customer information. Write what a person does from request to completion. Identify the point where a suggestion becomes an externally visible or difficult-to-reverse action.

Minutes 4–8 · Separate the verbs. List reading, drafting, editing and committing as different capabilities. Decide which are necessary for the first trial. Remove access that is merely convenient. Make the boundary understandable to a user without technical documentation.

Minutes 8–12 · Check the evidence. Specify which original record the user should inspect before accepting the output. Include a test where the newest instruction contradicts an older one. Decide how the assistant should display uncertainty or conflicting information.

Minutes 12–15 · Practise stopping. Run a safe example in which the task cannot be completed confidently. Check that work pauses and reaches a named person. Record the action taken and preserve the original files or messages for review.

Start with the real action boundary. Connect a useful assistant to a process your team can explain, so convenience does not quietly expand into authority that nobody intended to delegate.

Which action in our pilot changes a real commitment—and who must approve it?

Brought to you by Data Wave — your AI & Data Team as a Subscription.

Explore Data Wave