AI Leadership Weekly · Issue #65 · Tuesday 6 January 2026 · 08:00 GMT
Good morning.
NVIDIA opens the year with a new infrastructure roadmap. It is relevant to future capacity, but it is not a price cut on your current service. Alongside that announcement, this holiday edition revisits December’s updates to Claude Skills and Codex. The combination is a useful reminder to keep long-term technology planning separate from the small improvements a team can actually deliver this month.
IN 60 SECONDS
NVIDIA sets out Rubin. NVIDIA announced the Rubin platform and its deployment roadmap.
Skills become an organisational resource. Anthropic expanded how organisations can distribute and manage Skills, bringing reusable instructions into a more centrally managed setting.
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.
CEO / COO / CXO CHECKLIST
CEO: Separate future infrastructure from currently usable services.
COO: Turn useful instructions into maintained team assets.
CXO: Use coding assistance on a bounded change with a real acceptance test.
TOP STORIES
1. NVIDIA sets out Rubin
NVIDIA · 5 January 2026
What happened. NVIDIA announced its Rubin platform on 5 January, bringing together six new chips in a coordinated AI system. It said partner products would be available in the second half of 2026. Its claims about lower inference costs compare its platforms; they do not establish an immediate reduction in a customer’s software or cloud bill.
Why it matters. Our take: Keep infrastructure progress in the planning horizon without letting it replace the current business case. Ask when a capability will be available through your approved supplier, in the required location, with a service you can actually buy. An attractive future unit cost is not a reason to leave an expensive manual problem untouched today.
What to do. Put current and future options on separate lines of the roadmap. For the future option, record the expected availability, unresolved dependencies and decision trigger. For the current option, identify a small, reversible improvement that does not depend on a promised platform arriving on time.
2. December lookback: skills become an organisational resource
Anthropic · 18 December 2025
What happened. Anthropic’s 18 December release notes described organisation-wide skill management for Team and Enterprise plans, a partner skills directory and an open Agent Skills standard. Skills package repeatable ways of working. Those announcements do not establish that a shared instruction is correct for every team or remains correct after a process changes.
Why it matters. Our take: Treat a reusable skill like a maintained procedure, not a clever prompt with an unlimited shelf life. It should explain its purpose, inputs, expected output and limits. Assign someone to test changes and retire obsolete versions. A small collection of dependable instructions is more useful than a catalogue nobody is responsible for.
What to do. Choose one instruction a colleague already uses successfully. Turn it into a short team asset with an example, an owner and a version date. Have another person run it on an exception. Fix ambiguities before distributing it more widely or calling it a standard process.
3. December lookback: GPT-5.2-Codex reaches paid users
OpenAI · 18 December 2025
What happened. OpenAI introduced GPT-5.2-Codex on 18 December, describing improvements for longer software-engineering tasks, including refactoring and migrations. It began the rollout in Codex for paid ChatGPT users, with API access described as coming later. Its benchmark and security findings are supplier evaluations, not an assurance that generated changes are ready to release.
Why it matters. Our take: Start with maintenance work that has a clear finish: a tested correction, a documented upgrade or removal of a known source of support effort. Avoid the vague objective of building something impressive. Keep the person responsible for the service accountable for accepting and releasing the change, even when generation becomes easier.
What to do. Ask the engineering lead to nominate a small backlog item with existing tests. Run the work in an isolated branch, require a change explanation and review the result. Count the time spent specifying, testing and correcting it. Keep any production deployment outside the agent’s initial authority.
SIGNALS FROM THE LAST MONTH
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
16 December · ChatGPT updates image generation. OpenAI introduced a new ChatGPT Images experience and model, reporting improvements in editing and following detailed visual instructions. Source
11 December · GPT-5.2 launches. OpenAI released GPT-5.2, positioning the model for professional knowledge work, coding and longer, tool-assisted assignments. Source
9 December · Accenture partners with Anthropic. The companies announced a multiyear partnership combining Claude deployment, industry work and training for Accenture professionals. Source
IN BRIEF
More dated updates from the preceding 30 days.
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
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
17 December · Mistral upgrades document reading. Mistral introduced OCR 3, reporting better handling of scanned documents, handwriting and tables. Those claims still need task-specific testing. 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
THE 15-MINUTE PLAYBOOK
Build a six-week learning plan
Minutes 0–4 · Select one outcome. Choose a problem small enough for a named team to own. Describe the current friction and the improvement a recipient should notice. Keep the objective independent of a model brand or a future infrastructure release.
Minutes 4–8 · Choose a reusable asset. Identify what the trial should leave behind: an instruction, an acceptance test, a clean data set or a revised process. Give the asset an owner. Make its future usefulness part of the trial design.
Minutes 8–12 · Set the comparison. Record the existing method and the cases to test. Include an awkward example and the effort needed to review results. Decide how to capture evidence without making the measurement exercise larger than the work itself.
Minutes 12–15 · Book the decision. Set a review point with three possible outcomes: expand, change the approach or stop. State the evidence needed for each. Keep future technology dependencies visible, but do not let them prevent a reversible learning step now.
DATA WAVE MOMENT
Start the year with a practical learning loop. Improve one piece of work, preserve what the team learns and make the next funding decision on evidence rather than the size of the roadmap.
QUESTION FOR READERS
What useful evidence and reusable asset will our next six weeks of AI work produce?
Brought to you by Data Wave — your AI & Data Team as a Subscription.
