AI Leadership Weekly · Issue #98 · Tuesday 25 August 2026 · 08:00 BST
Good morning.
A useful AI feature does not always announce itself with a new chat window. This week’s examples sit inside an experience somebody already understands: following a tennis match, working through a consultation or returning to something on a computer.
That makes the benefits easier to imagine—and the boundaries more important to describe. The right context for one person can be excessive access for another. Our focus this week is therefore the fit between the feature and the moment it is meant to help. Start there, rather than with the longest list of integrations.
IN 60 SECONDS
Experience: IBM and the USTA add AI-powered ways to follow the US Open. Workflow: Microsoft opens an app route around Dragon Copilot’s physician experience in the US. Context: ChatGPT expands a desktop-history feature to the UK. These are three separate examples, with different users, eligibility and responsibilities—not one universal workplace rollout.
CEO / COO / CXO CHECKLIST
CEO: Ask which user need the proposed feature serves at a particular moment.
COO: Observe where a person leaves the main workflow to find information or finish a task.
CXO: Make permissions, optional features and unsupported regions visible before starting a trial.
TOP STORIES
1. IBM and the USTA build AI into the fan experience
IBM / USTA · 24 August 2026
What happened. On 24 August, IBM and the USTA announced AI-powered US Open features including personalised updates, match context and interactive experiences. The announcement describes product features for following the tournament, not evidence that the same approach will improve every customer service.
Why it matters. Our take: the interesting design choice is to organise information around what the user is doing now. A fan following a match does not need an undifferentiated dump of tennis data. The same question is useful elsewhere: what does a customer or colleague need at this point in the journey? A small, timely feature may be more helpful than a broad assistant with no clear job.
What to do. Choose one customer touchpoint. List the two questions people most often need answered there and the approved information required. Sketch a focused feature that addresses those questions without making the user leave the journey or ask for a complete briefing.
2. Dragon Copilot gets a route for physician-workflow apps
Microsoft, 20 August entry · 20 August 2026
What happened. Microsoft’s 20 August partner announcement describes a Marketplace route for apps within Dragon Copilot’s physician workflow. The release is US-only. It should not be presented as an available NHS deployment or as evidence of improved clinical outcomes.
Why it matters. Our take: the wider business idea is to put a useful function where the person already works. An extra application is not automatically helpful when it creates more copying, switching and re-entry. In specialist settings, suitability also depends on professional oversight, information handling and the exact supported workflow. Familiar software is an entry point for assessment, not a waiver of those requirements.
What to do. Map one handoff between your main application and a supporting tool. Count the information copied across and identify the decision made there. Ask whether an embedded feature could remove that handoff, while retaining the checks and responsibilities that make the process acceptable.
3. ChatGPT’s computer-history feature reaches the UK
OpenAI, 20 August entry · 20 August 2026
What happened. OpenAI’s 20 August notes expand Computer History for eligible Pro users on macOS to the UK, EEA and Switzerland. It requires Memories and is off by default. Availability does not amount to permission to enable it on a managed work computer.
Why it matters. Our take: remembering context and having permission to retain it are different questions. A person may find continuity helpful while their employer needs limits on what information is captured and how it is handled. Avoid both extremes: assuming every contextual feature is harmless, or treating the ability to remember anything as inherently unusable. Review the actual feature and the intended setting.
What to do. Before enabling it for work, have the appropriate internal owner review its data handling and approved uses. Give employees an understandable explanation of the choice. Keep it disabled where approval is absent; a useful experiment does not require bypassing workplace controls.
SIGNALS FROM THE LAST MONTH
13 August: The IBM–OpenAI partnership makes implementation capacity part of the enterprise buying discussion. IBM / OpenAI · 13 August 2026
13 August: Another Gemini Flash release reinforces the value of a repeatable upgrade test. Google · 13 August 2026
6 August: Apptio’s preview connects AI economics to the need for a defensible baseline. IBM · 6 August 2026
29 July: IBM’s study remains useful context, provided its breached-organisation sample and limits stay attached. IBM / Ponemon Institute · 29 July 2026
IN BRIEF
Further dated updates, including recent context worth keeping in view.
24 August: Microsoft announces a personalised 30-day Copilot trial-planning tool. Microsoft, 24 August entry · 24 August 2026
20 August: OpenAI adds controls for changing supported hosted-site URLs; this is separate from custom-domain support. OpenAI, 20 August entry · 20 August 2026
11 August lookback: IBM’s new Together AI capacity is planned for 2027, not available on announcement day. IBM / Together AI · 11 August 2026
10 August lookback: Google’s Ads and Analytics updates remain a mix of beta and forthcoming features. Google · 10 August 2026
THE 15-MINUTE PLAYBOOK
Minutes 0–4: Choose a real user and one moment in their working day. Avoid designing for “every employee” at once.
Minutes 4–8: Write the question or task they are trying to resolve. Note where the answer should come from and what they do next.
Minutes 8–12: Draw the smallest useful intervention in the existing workflow. Include how the user checks the answer and what happens when it cannot be produced.
Minutes 12–15: Ask somebody who does the work to challenge the sketch. Would it remove a step, or merely place a new panel beside the old steps? Capture their objection before commissioning the build. The most valuable feedback may be that the proposed feature solves the wrong moment.
DATA WAVE MOMENT
At Data Wave, “build from within” means taking the user’s working context seriously. The opportunity might be a new product, an embedded feature or simply removing a handoff that no longer needs to exist. The important part is whether it makes the next action clearer and the overall service better—not whether it adds another place to type a prompt.
QUESTION FOR READERS
Where in your customer or employee journey would a small, well-timed AI feature be more useful than a general assistant?
Brought to you by Data Wave — your AI & Data Team as a Subscription.
