AI Leadership Weekly · Issue #83 · Tuesday 12 May 2026 · 08:00 BST
Good morning.
Uber’s use of AI, Google’s AlphaEvolve update and Microsoft’s analysis of working patterns all look beyond a standalone chat assistant. They focus on decisions, optimisation and how work is organised. This edition keeps the examples separate from the conclusions: a customer account can show a promising pattern, but your own process still needs a clear problem and a way to measure improvement.
IN 60 SECONDS
Microsoft studies new working patterns. Microsoft described different patterns of collaboration between employees and agents, focusing on the organisation of work rather than a single product launch.
Uber describes its AI assistant. An OpenAI customer account described Uber’s use of AI to connect assistance with the information and decisions involved in operating its services.
AlphaEvolve reports further applications. Google described additional uses of AlphaEvolve for algorithmic optimisation, extending AI applications beyond conversational assistance.
CEO / COO / CXO CHECKLIST
CEO: Choose a repeated decision and observe how somebody makes it today.
COO: Identify whether the missing ingredient is information, judgement, calculation or authority.
CXO: Measure the quality and effort of the decision, not only use of the interface.
TOP STORIES
1. Microsoft studies new working patterns
Microsoft · 5 May 2026
What happened. Microsoft’s 5 May Work Trend Index article combined product-use analysis with a survey of 20,000 AI-using workers across ten countries. It described different patterns of collaboration, from assistance with individual work to directing more complete tasks. The survey concerns AI users, not the entire workforce, and the findings do not prove causation.
Why it matters. Our take: Do not make maximum autonomy the goal for every job. A colleague may need to retain the reasoning because the decision is part of their expertise. Elsewhere, repeatedly constructing the same first draft may add little value. Discuss the work in specific chunks and decide what the person should still produce, inspect and approve. This is more useful than labelling a whole role “automatable”.
What to do. Ask a team to identify one task to keep doing themselves, one to draft with help and one to delegate within limits. Explain the reason for each choice before selecting tools.
2. Uber describes its AI assistant
OpenAI / Uber · 6 May 2026
What happened. OpenAI published a 6 May account of Uber’s AI work, including an assistant that explains marketplace information to drivers. The case describes routing requests to different specialist systems and using different model sizes for different tasks. It is a supplier-published customer example, not independently measured evidence of improved earnings.
Why it matters. Our take: The transferable idea is to reduce the effort of interpreting a complex situation at the point of work. A helpful frontline assistant should know what question it is answering, which information is current and what authority the user retains. Avoid building a broad conversational layer when a small number of well-supported questions would address the real need.
What to do. Interview three frontline colleagues about a repeated decision. Build a sample response that shows its source, explains the options and leaves the decision with the appropriate person. Test whether it is useful in context.
3. AlphaEvolve reports further applications
Google · 7 May 2026
What happened. Google’s 7 May AlphaEvolve update described applications of algorithmic search in its infrastructure and customer work, including supply chains and warehouse design. It also discussed scientific and simulation results. Those categories should remain separate: a simulated improvement is not the same as a proven operational deployment.
Why it matters. Our take: Some valuable problems have a clear objective and constraints: reduce wasted capacity, improve a schedule or explore more feasible designs. For those tasks, the output may be a better algorithm or plan rather than a written answer. Start with a trusted evaluator and a baseline. Without those, a system can optimise the wrong measure extremely efficiently.
What to do. Ask operations for one constrained planning problem. Write the objective, non-negotiable limits and current baseline. Decide how an improved proposal would be tested before it affected live work.
SIGNALS FROM THE LAST MONTH
23 April · Managed Agents gains memory. Anthropic added inspectable, persistent memory for Claude Managed Agents, letting developers manage information carried between sessions. Source
22 April · Google introduces an agent platform. Google announced Gemini Enterprise Agent Platform, bringing development, deployment and management of enterprise agents into a connected offering. Source
21 April · Google updates Deep Research. Google announced its next generation of Gemini Deep Research, extending its developer offering for multi-step research tasks. Source
20 April · Google AI subscriptions connect to AI Studio. Google introduced a route for eligible AI subscribers to prototype in AI Studio, subject to the service’s usage conditions. Source
IN BRIEF
More dated updates from the preceding 30 days.
4 May · IBM surveys changing leadership roles. IBM’s study of 2,000 CEOs and equivalent leaders reported changes in AI responsibilities; its results describe that surveyed population. Source
28 April · IBM Bob becomes generally available. IBM launched Bob globally as an AI development partner spanning planning, coding, testing, deployment and modernisation. Source
28 April · Microsoft showcases customer deployments. Microsoft published a roundup of customer AI projects, describing how businesses were connecting company information with day-to-day work. Source
23 April · GPT-5.5 launches. OpenAI introduced GPT-5.5, broadening its model offering for professional and coding work; API availability followed the initial announcement. Source
THE 15-MINUTE PLAYBOOK
Minutes 0–4: Describe a repeated decision, its owner and the point at which it is made. Write down what a good outcome looks like.
Minutes 4–8: List the information required and where it comes from. Separate current facts from assumptions and established rules from personal preferences.
Minutes 8–12: Choose a proposed intervention: better information, a prepared option, a calculation or a bounded automated step. Explain why it fits the obstacle.
Minutes 12–15: Define a small comparison with the current method. Include time, correctness, user understanding and exceptions. Agree what would justify continuing rather than treating use of the prototype as success in itself.
DATA WAVE MOMENT
Start with the decision someone is trying to make, then build the information and capability around it. That approach can reveal opportunities in service, planning and operations that a generic assistant rollout misses. The objective is useful work in context, not a new screen that people must learn to navigate.
QUESTION FOR READERS
Which important decision is still difficult mainly because the right information arrives in the wrong form?
Brought to you by Data Wave — your AI & Data Team as a Subscription.
