AI Leadership Weekly · Issue #62 · Tuesday 16 December 2025 · 08:00 GMT

Good morning.

A new OpenAI model leads the week, but the surrounding announcements matter too. The Linux Foundation is giving agent projects a shared organisational home, while Accenture and Anthropic are expanding deployment and training work. Capability, compatibility and delivery skills are three different purchases. A credible AI plan needs to explain how they will come together, rather than treating the latest model as the whole answer.

GPT-5.2 launches. OpenAI released GPT-5.2, positioning the model for professional knowledge work, coding and longer, tool-assisted assignments.

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.

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

https://openai.com/index/introducing-gpt-5-2/

OpenAI · 11 December 2025

What happened. OpenAI introduced GPT-5.2 on 11 December, describing improvements in areas including spreadsheets, presentations, long-context work and tool use. Its announcement covered API availability and a staged ChatGPT rollout. The professional-task comparisons are evaluation results reported by the supplier, not a forecast of how many roles a customer can remove.

Why it matters. Our take: Test the deliverable a colleague actually needs. An attractive presentation may still misstate the source analysis; a plausible spreadsheet may still contain the wrong assumptions. Put the assessment at the point where work is accepted, with checks that reflect the consequence of an error. Do not confuse format with completion.

What to do. Choose a recurring briefing or analysis pack and give the new option the same approved source material as the existing method. Ask the recipient to assess traceability, accuracy and usability. Record review time and correction work, not just the time needed to create a first draft.

https://www.linuxfoundation.org/press/linux-foundation-announces-the-formation-of-the-agentic-ai-foundation

Linux Foundation · 9 December 2025

What happened. The Linux Foundation announced the Agentic AI Foundation on 9 December, with founding contributions including Anthropic’s Model Context Protocol, Block’s goose and OpenAI’s AGENTS.md. The initiative brings open agent-related projects into a shared governance setting. Membership or standards support is not, by itself, evidence that products can be swapped without rework.

Why it matters. Our take: Ask for a practical portability demonstration. Can the team retain its task instructions, tests and connection definitions while replacing a component? Separate the parts governed by a standard from supplier-specific features. You do not need perfect portability everywhere, but you should understand where a dependency becomes expensive to leave.

What to do. Choose one small workflow and ask the delivery team to mark its reusable and supplier-specific elements. Put acceptance tests outside the supplier demonstration. Record what a second implementation would require, including permissions and operational support, before describing the design as interchangeable.

https://www.anthropic.com/news/anthropic-accenture-partnership

Anthropic · 9 December 2025

What happened. Anthropic and Accenture announced a partnership on 9 December, including an Accenture Anthropic Business Group and plans to train approximately 30,000 Accenture professionals on Claude. The programme also described work on deployment and value measurement. Planned training is not the same as completed training or independently verified customer improvement.

Why it matters. Our take: Make learning specific to a role and a task. A course can explain a tool, but your organisation still needs to show what a good result looks like and when an employee should stop. Involve the person who signs off the work, not only the person expected to use the assistant.

What to do. Replace one generic awareness session with a worked example from the team’s real process. Include an incomplete input, a plausible error and an escalation. Assess whether someone can produce and review an acceptable output, rather than whether they attended or enjoyed the session.

26 November · How long-running agents hand over. Anthropic described an engineering approach combining environment setup, progress records and incremental work to help coding agents continue across sessions. Source

25 November · MCP marks its first year. The Model Context Protocol published its November specification update, including additions for authorisation and longer-running work. Source

24 November · Opus 4.5 launches. Anthropic released Opus 4.5 with API launch pricing of $5 per million input tokens and $25 per million output tokens. Source

24 November · Claude expands tool use. Anthropic introduced tool search, programmatic tool calling and tool-use examples for developers building agents with larger collections of tools. Source

More dated updates from the preceding 30 days.

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

2 December · Amazon expands Nova. Amazon announced new Nova models and Nova Forge, an offering for organisations seeking deeper customisation during model development. Source

2 December · Mistral 3 offers open models. Mistral introduced its third model generation, spanning smaller models and Mistral Large 3 under an Apache 2.0 licence. Source

1 December · DeepSeek updates its model line. DeepSeek released V3.2 and a separate Speciale variant. The latter’s temporary endpoint had different capabilities and availability conditions. Source

Minutes 0–4 · Pick the work product. Choose a document, decision pack or service action with a clear recipient. Write the acceptance criteria in ordinary language. Name the expert who can judge whether it is fit for purpose and why it matters.

Minutes 4–8 · Map the dependencies. List the approved data, model, connections and instructions used to produce it. Identify the supplier-specific elements. Ask what would need to change if the preferred model were unavailable for a week.

Minutes 8–12 · Design the learning exercise. Build one representative example and one difficult exception. Have the user and reviewer work through both. Record what each role needs to understand, including how to recognise uncertainty and request help.

Minutes 12–15 · Set one joint decision. Bring the process owner, delivery lead and learning lead together. Agree the next trial, the evidence to collect and the person who can approve expansion. Avoid three separate sign-offs that never examine the whole service.

Build capability around a real piece of work. Keep the model choice, the learning exercise and the evidence of value connected, so improvements survive beyond a single enthusiastic demonstration.

Who is responsible for joining our model, process and skills decisions into one deliverable?

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

Explore Data Wave