AI Leadership Weekly · Issue #91 · Tuesday 7 July 2026 · 08:00 BST
Good morning.
Microsoft is putting people behind implementation, OpenAI is bringing ChatGPT into PowerPoint, and Anthropic has launched Sonnet 5. The three announcements address different obstacles: delivery capacity, awkward movement between tools and the cost of model capability. This edition looks at which obstacle each could remove. The useful measure is the finished piece of work, including checking and correction—not the speed of the first draft.
IN 60 SECONDS
Microsoft funds hands-on implementation. Microsoft announced a $2.5 billion initiative involving 6,000 industry and engineering experts to work alongside customers on AI implementation.
ChatGPT comes to PowerPoint. OpenAI made ChatGPT for PowerPoint generally available, offering in-application assistance with editable presentations, subject to account limits and workspace settings.
Sonnet 5 arrives. Anthropic launched Sonnet 5 with introductory API pricing of $2 per million input tokens and $10 per million output tokens.
CEO / COO / CXO CHECKLIST
CEO: Choose one existing deliverable whose preparation time is worth reducing.
COO: Count checking and correction time, not only first-draft speed.
CXO: Separate implementation support from software and model costs.
TOP STORIES
1. Microsoft funds hands-on implementation
Microsoft · 2 July 2026
What happened. On 2 July, Microsoft announced Microsoft Frontier Company, backed by a $2.5 billion investment and 6,000 industry and engineering experts to work alongside customers. The offer combines implementation, industry knowledge and change management.
Why it matters. My reading is that delivery capacity is becoming part of the product. For buyers, that creates an opportunity to demand more than access to tools and a promising demonstration.
A useful supplier proposal should explain what will improve, what your team must contribute, and what capability stays with you afterwards. Buying help should not mean renting the ability to understand your own business.
What to do. Take one live AI proposal and replace its activity measures with a business result: faster turnaround, fewer avoidable errors, or more completed work. Ask who is responsible for achieving it.
2. ChatGPT comes to PowerPoint
OpenAI release notes · 6 July 2026
What happened. OpenAI made ChatGPT for PowerPoint generally available globally on 6 July. Its sidebar can create and revise presentations while keeping slides editable. Access spans ChatGPT plans, subject to limits and workspace settings; advanced edits and template matching still require review.
Why it matters. This offers a practical alternative to copying text out of a chat and rebuilding the presentation yourself. The important test is the time saved getting to a usable deck—not how quickly a first version appears.
There is also a useful discipline here: decide the story before asking for the slides. A presentation can be beautifully formatted and still avoid the decision everybody came to make.
What to do. Try it on one existing presentation with an agreed message and trusted figures. Compare the full preparation and correction time with your usual approach.
3. Sonnet 5 arrives
Anthropic · 30 June 2026
What happened. Anthropic launched Claude Sonnet 5 on 30 June across its plans and API. The company reports better coding, tool use and multi-step performance than Sonnet 4.6, with some results approaching Opus 4.8. Launch API pricing was $2 per million input tokens and $10 per million output tokens. These are supplier-reported comparisons, not a forecast of your savings.
Why it matters. The commercial question is whether a less expensive model can finish your particular task well enough. Paying for maximum capability on every routine job may be unnecessary; buying the cheapest response can also be a false economy.
What to do. Re-run a small set of completed, representative jobs. Compare accepted results, corrections and total cost. Keep the difficult cases in the sample—they are the ones most likely to change the answer.
SIGNALS FROM THE LAST MONTH
16 June · Apptio previews conversational spend analysis. Apptio announced Conversational Insights and other AI capabilities for technology-spend analysis, with the conversational experience initially in preview. Source
16 June · Microsoft sets out its AI approach. Microsoft published its perspective on model diversity, enterprise control and implementation; it was a strategy article, not a new model release. Source
11 June · IBM and ServiceNow expand collaboration. The companies announced work connecting enterprise data and workflows, with initial new offerings planned for the second half of 2026. Source
9 June · Gemini expands live translation. Google introduced Gemini Live 3.5 Translate, broadening speech-translation capabilities while retaining product-specific availability conditions. Source
IN BRIEF
More dated updates from the preceding 30 days.
24 June · Gemini 3.5 Flash adds computer use. Google announced computer-use capabilities for Gemini 3.5 Flash, extending how the model can interact with software interfaces. Source
23 June · Mistral releases OCR 4. Mistral introduced OCR 4, adding richer structural information and paragraph-level locations to extracted document content. Source
23 June · Azure adds an Observability Agent. Microsoft announced general availability of Azure’s Observability Agent, applying agents to the investigation and operation of cloud services. Source
18 June · Claude Code adds interactive artifacts. Anthropic introduced artifacts in Claude Code in beta, allowing users to create and share interactive outputs from their development work. Source
THE 15-MINUTE PLAYBOOK
Choose a repeatable deliverable: a customer proposal, a management pack, a briefing or an analysis.
Minutes 0–4: Write down what makes it useful. Name the audience, the decision and the information that must be correct. Find one previously accepted example.
Minutes 4–8: Break the work into gathering, thinking, producing and checking. Estimate where the effort goes; label estimates honestly rather than dressing them up as measurements.
Minutes 8–12: Pick the one step to test with AI. Decide how you will time it and what would count as a failed result. Use approved, appropriate information.
Minutes 12–15: Assign a person and a real deadline. Review the finished output together, including correction time. Only then decide whether to repeat, change or stop the experiment.
The point is not to launch an AI programme in fifteen minutes. It is to set up a test that tells you something useful.
DATA WAVE MOMENT
Start with the work that needs doing, not a shopping list of tools.
For an overloaded team, the useful question is often: which part should become quicker, which part needs better information, and which part should disappear altogether? That is the conversation worth having before the next licence purchase.
QUESTION FOR READERS
What would you do with the time saved—and would anybody notice the improvement?
Brought to you by Data Wave — your AI & Data Team as a Subscription.
