AI Leadership Weekly · Issue #63 · Tuesday 23 December 2025 · 08:00 GMT

Good morning.

Google’s Gemini 3 Flash and OpenAI’s image update both put more emphasis on producing usable work quickly. A proposed Coursera–Udemy combination brings the learning side into view. For leaders, this is a useful final pre-Christmas check on the difference between producing more material and helping people do better work. A bigger pile of drafts is not quite the same Christmas present.

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

ChatGPT updates image generation. OpenAI introduced a new ChatGPT Images experience and model, reporting improvements in editing and following detailed visual instructions.

Coursera and Udemy plan to combine. Coursera and Udemy announced a proposed combination.

https://blog.google/products-and-platforms/products/gemini/gemini-3-flash/

Google · 17 December 2025

What happened. Google introduced Gemini 3 Flash on 17 December, positioning it as a faster, more cost-efficient member of the Gemini 3 family. It also announced product and developer access, including a change to the default model in the Gemini app. Claims about performance and efficiency are Google’s comparisons, not guaranteed results for a particular workflow.

Why it matters. Our take: Speed matters most where waiting interrupts the person doing the work. It matters less when a task runs overnight and requires substantial review in the morning. Segment the workload before setting a universal model policy. Keep the quality threshold fixed, then ask where a faster response creates a meaningful service improvement.

What to do. Choose one interactive task and one batch task. Compare elapsed time, corrections and accepted results using the same examples. Record the product setting as well as the model name. Keep a route for difficult exceptions instead of making the fastest option handle every case.

https://openai.com/index/new-chatgpt-images-is-here/

OpenAI · 16 December 2025

What happened. OpenAI released its new ChatGPT Images model and GPT Image 1.5 in the API on 16 December. The announcement emphasised instruction following, editing and preserving details during changes, while acknowledging remaining limitations. It distinguished model access from the separate Images experience, whose Business and Enterprise rollout was described as coming later.

Why it matters. Our take: Treat visual generation as part of a production process, not as an automatic publishing decision. Decide which elements may change and which must be exact: a product shape, label, logo or offer. A convincing image can still be unsuitable for its intended use. Keep the person accountable for the final asset involved.

What to do. Build a small approved brief using a product or campaign your team knows well. Request variants, then inspect wording, consistency and unintended changes at full size. Retain the source assets and revision history. Require the normal brand and content approval before an image reaches customers.

https://investor.coursera.com/events-and-presentations/event-details/2025/Coursera-to-combine-with-Udemy-2025-S7iaPqZ8al/default.aspx

Coursera · 17 December 2025

What happened. Coursera and Udemy announced a definitive agreement on 17 December to combine in an all-stock transaction. Their investor communication describes a proposed merger, with forward-looking expectations subject to uncertainty. It should not be read as a completed integration or a promise that existing learning products, pricing and contracts have already changed.

Why it matters. Our take: Use the announcement as a prompt to review your learning approach, not to pause useful training until a corporate transaction resolves. Decide which skills your teams need and what successful application looks like. A larger course catalogue does not tell you whether a colleague can safely complete the job you are changing.

What to do. Review one AI learning pathway before renewing or expanding it. Identify the role, the practice task and the manager who will observe application. Keep supplier-contract questions separate from learning-design questions. Preserve access to the material people currently need while any platform changes remain proposals.

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 · A foundation for agent standards. The Linux Foundation launched the Agentic AI Foundation, bringing projects including MCP, goose and AGENTS.md into a shared organisational home. Source

9 December · Accenture partners with Anthropic. The companies announced a multiyear partnership combining Claude deployment, industry work and training for Accenture professionals. Source

8 December · OpenAI surveys enterprise adoption. OpenAI published enterprise usage and survey findings. Reported time savings describe its study population, not a guaranteed return for other businesses. Source

More dated updates from the preceding 30 days.

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 · Skills become an organisational resource. Anthropic expanded how organisations can distribute and manage Skills, bringing reusable instructions into a more centrally managed setting. 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 · 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

Minutes 0–4 · Choose the recipient. Name who will use the output and what they must be able to do with it. Pick a single format, such as a customer briefing or campaign image. State the details that must remain exact.

Minutes 4–8 · Set the quality bar. Write three checks before generating anything. Include one that requires a human judgement and one that can be verified against source material. Add the circumstances in which the output must not be published or used.

Minutes 8–12 · Count the full journey. Time preparation, generation, selection, review and correction separately. Keep rejected variants in the record. Do not report a quick first draft as a completed business task if the recipient still needs substantial rework.

Minutes 12–15 · Teach the workflow. Have another colleague repeat the exercise using the same instructions. Ask where the brief or review checklist is unclear. Update those artefacts before buying more capacity or making a wider rollout commitment.

Keep production and quality in the same conversation. Improve the brief, the evidence and the handover together, so a faster first draft becomes a genuinely easier job for the next person.

Are we removing work, or simply sending more drafts to the people who must approve them?

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

Explore Data Wave