AI Leadership Weekly · Issue #56 · Tuesday 4 November 2025 · 08:00 GMT
Good morning.
This week’s announcements sit behind the tools as much as inside them. GitHub introduced Agent HQ, Microsoft and OpenAI revised their relationship, and OpenAI announced a major AWS compute agreement. The names overlap, but the promises do not: buying infrastructure, distributing models and managing coding agents are separate parts of the story. Here is what each announcement actually changes.
IN 60 SECONDS
GitHub introduces Agent HQ. GitHub announced a common environment for directing coding agents, with broader third-party agent integrations planned over subsequent months.
Microsoft and OpenAI revise terms. The partners announced a revised agreement, including additional Azure purchasing commitments and changes to compute sourcing, not unrestricted API distribution.
AWS supplies OpenAI compute. OpenAI announced a seven-year, $38 billion AWS agreement.
CEO / COO / CXO CHECKLIST
CEO: Measure reviewed, accepted work rather than the volume of agent activity.
COO: Record the exact contracted service, not just the model or supplier brand.
CXO: Test how the workflow continues when its preferred AI service is unavailable.
TOP STORIES
1. GitHub introduces Agent HQ
GitHub · 28 October 2025
What happened. On 28 October, GitHub introduced Agent HQ: a plan to bring coding agents into a shared workflow, with a central view for assigning and tracking work. GitHub said agents from additional providers would become available over the coming months; the announcement was not evidence that every integration was already live.
Why it matters. Our take: Ask what happens after an agent produces a change. Your acceptance decision still needs tests, a reviewer and an owner. More simultaneous attempts are only useful when the team can distinguish work worth keeping from work that needs correction.
What to do. Agree one bounded task for an agent-assisted trial. Set a limit on concurrent work, record review effort and count changes accepted without substantial rework. Keep the existing approval point for anything entering a live business service.
2. Microsoft and OpenAI revise terms
Microsoft · 28 October 2025
What happened. Microsoft and OpenAI announced a revised agreement on 28 October. Microsoft said OpenAI had committed to an additional $250 billion of Azure services and that Microsoft no longer held first-refusal rights over compute supply. Its statement also retained Azure API exclusivity, subject to the agreement’s conditions and exceptions.
Why it matters. Our take: Separate the infrastructure a model supplier uses from the service your business actually buys. Do not infer a new deployment option, data location or exit route from a partnership headline. Ask for confirmation that applies to your particular product and contract.
What to do. Take one AI purchase through a simple chain: the business application, the service provider, the model, the agreed processing locations and the support owner. Mark unanswered questions explicitly rather than filling them with assumptions about supplier relationships.
3. AWS supplies OpenAI compute
OpenAI · 3 November 2025
What happened. On 3 November, AWS and OpenAI announced a $38 billion, seven-year agreement for compute supporting training and inference. The announcement described immediate use and a target to deploy the full capacity before the end of 2026. It did not announce general access to proprietary OpenAI APIs through Amazon Bedrock.
Why it matters. Our take: Read this as a capacity and supplier-development signal, not a promise of lower prices or better availability for your account. Business continuity should rest on your own service commitments and a tested fallback, rather than on announced future infrastructure.
What to do. Walk through a day without one AI service. Decide which tasks can queue, which need a manual method and which have an approved alternative. Check that the alternative can actually receive the necessary information and produce an acceptable result.
SIGNALS FROM THE LAST MONTH
16 October · Copilot expands on Windows. Microsoft announced more voice and screen-aware Copilot experiences. The experimental Actions capability was still headed towards Windows Insiders. Source
15 October · A smaller Claude arrives. Anthropic launched Haiku 4.5, with API launch pricing of $1 per million input tokens and $5 per million output tokens. Source
13 October · OpenAI plans new accelerator capacity. OpenAI and Broadcom announced a 10-gigawatt collaboration, targeting deployment from the second half of 2026 through 2029. Source
9 October · Google launches Gemini Enterprise. Google introduced a workplace platform connecting company information, applications and agents, with tools for building and deploying additional agents. Source
IN BRIEF
More dated updates from the preceding 30 days.
27 October · Claude enters Excel in preview. Anthropic announced an Excel research preview and expanded financial-services capabilities. Access was limited rather than generally available. Source
23 October · Company knowledge in ChatGPT. OpenAI introduced connected company knowledge for Business, Enterprise and Edu, with source citations and existing access permissions. Source
21 October · ChatGPT gets a browser. OpenAI launched Atlas on macOS, with browser-agent functionality in preview for eligible paid plans rather than universal access. Source
16 October · Claude gains reusable Skills. Anthropic introduced Skills: packages of instructions and resources that Claude can use for recurring tasks, including document work. Source
THE 15-MINUTE PLAYBOOK
Map one AI dependency from outcome to recovery
Minutes 0–4: Start with the business consequence. Select one AI-assisted workflow. Define the result the business needs, when it is needed and what happens if it is wrong or late.
Minutes 4–8: Name the services and owners. Identify the application, contracted provider, model service, information owner and person responsible for support. Mark anything that still needs verification.
Minutes 8–12: Describe the failure response. Choose what to do when the assistant is unavailable or its output fails review: queue the work, complete it manually or use an approved alternative. Note the information and skills that response requires.
Minutes 12–15: Set a test and a decision. Assign someone to rehearse the fallback, record the effort and report the result. Decide whether the evidence supports wider use, a smaller scope or a pause.
The output is a one-page responsibility map with a usable fallback, not a diagram of supplier logos.
DATA WAVE MOMENT
Build AI into a business process that somebody owns. Connect delivery, data, controls and support to the outcome you want, then test the whole path — including what happens when the preferred approach does not work.
QUESTION FOR READERS
Which AI dependency would interrupt real work tomorrow — and does its owner have a tested alternative?
Brought to you by Data Wave — your AI & Data Team as a Subscription.
