AI Leadership Weekly · Issue #87 · Tuesday 9 June 2026 · 08:00 BST
Good morning.
This week is as much about where AI runs as what it can do. Microsoft outlined its connected enterprise platform, Google added Gemma 4 12B, and Anthropic announced support for Apple’s Foundation Models framework. Local, cloud and managed-platform options carry different practical consequences. The useful architecture discussion starts with the information and response the business needs, not a preference for one deployment label.
IN 60 SECONDS
Microsoft connects its agent stack. Microsoft outlined an integrated enterprise agent platform, emphasising the systems needed to run AI alongside existing business operations.
Gemma 4 gains a 12B model. Google introduced Gemma 4 12B, adding another option for organisations considering local or self-managed inference.
Claude connects with Apple’s framework. Anthropic announced Claude support for Apple’s Foundation Models framework, providing a route beyond the framework’s on-device models.
CEO / COO / CXO CHECKLIST
CEO: Map every place information travels during one real task.
COO: State what the user should experience when the network or model service fails.
CXO: Give each component a clear owner and a practical recovery route.
TOP STORIES
1. Microsoft connects its agent stack
Microsoft · 2 June 2026
What happened. At Build on 2 June, Microsoft described an integrated approach spanning development, enterprise context, governance and operations, connecting products including GitHub, Azure, Foundry and its business platforms. This was a platform direction with announced capabilities, not evidence that every possible component or customer configuration was already generally available.
Why it matters. Our take: Integration can remove repeated setup and make operating records easier to connect. The buying decision should still begin with the task you need to support. Ask how a request moves between components, where policy is enforced and how a failed step is recovered. Then check whether the result remains understandable to the people running the service, rather than only to specialists in the supplier’s stack.
What to do. Request a demonstration of one complete workflow with a deliberate failure. Follow its identity, records, evidence and recovery. Use the same scenario when evaluating an alternative design.
2. Gemma 4 gains a 12B model
Google · 3 June 2026
What happened. Google released Gemma 4 12B on 3 June, describing a model that can run locally on hardware with 16GB of suitable memory and handle text, vision and audio inputs. It is an additional member of the open Gemma family, not a feature that was already part of April’s initial release.
Why it matters. Our take: Local processing may be worth testing for bounded workloads where responsiveness, connectivity or information movement matters. But inspect the surrounding application too: logging, backups, external tools and updates may still contact other services. The operational comparison should include device management and the experience of users with different hardware. A successful demonstration on one laptop does not establish a fleet-wide deployment plan.
What to do. Pick a modest task and test it on representative devices using approved sample data. Measure responsiveness and output quality, and document any external service the application still needs.
3. Claude connects with Apple’s framework
Anthropic · 8 June 2026
What happened. Anthropic announced Claude support for Apple’s Foundation Models framework on 8 June, with availability stated for the following day. It described a way for developers to combine Apple’s on-device approach with Claude’s cloud capabilities. As of this edition’s cutoff, it was an announced developer route—not confirmed live access for every user or on-device Claude.
Why it matters. Our take: A hybrid design should make escalation understandable. What information leaves the device, why is the cloud step needed and what happens when that step is unavailable? Keep those decisions visible in the product design. A seamless interface should not mean an invisible change in where information is processed or which service is making the decision.
What to do. Write a sample user journey with one local step and one cloud step. Ask a non-technical colleague to explain where their information goes. Revise the design where the explanation is unclear.
SIGNALS FROM THE LAST MONTH
19 May · Managed Agents adds deployment choices. Anthropic introduced self-hosted sandboxes and MCP tunnels, allowing organisations to separate aspects of execution from managed orchestration. Source
19 May · Google develops a universal shopping cart. Google announced a universal shopping-cart experience, extending its work on AI-assisted discovery and purchasing. Source
19 May · Gemini 3.5 Flash launches. Google introduced Gemini 3.5 Flash while describing Pro as forthcoming, making the distinction between released and planned models important. Source
15 May · Databricks tests enterprise agent workloads. Databricks described GPT-5.5 results on enterprise document tasks. The published comparisons were supplier-reported evaluations, not customer-wide guarantees. Source
IN BRIEF
More dated updates from the preceding 30 days.
28 May · Endava describes Codex adoption. An OpenAI case study described how Endava was using Codex and reusable expertise in its delivery work; outcomes were customer-reported. Source
28 May · Mistral launches Search Toolkit. Mistral introduced components for building search and retrieval, giving developers more control over how agents find relevant information. Source
27 May · Mistral moves into physics AI. Mistral introduced a physics AI initiative, extending its focus from language models towards scientific and engineering applications. Source
22 May · Mistral introduces remote agents. Mistral added remote agents to Vibe and released Medium 3.5, widening its offer for delegated development work. Source
THE 15-MINUTE PLAYBOOK
Minutes 0–4: Pick a task and draw its user device, model, source systems and destination. Keep the picture simple enough for the business owner to follow.
Minutes 4–8: Mark the information moving between them, including logs and intermediate outputs. Identify any movement that needs a separate decision.
Minutes 8–12: Remove one connection from the diagram. Describe what the user should see and whether the work can pause, continue safely or move to a person.
Minutes 12–15: Add the support owner for each component and the route for recovering an incomplete task. Use the diagram as a shared design record before adding more integrations.
DATA WAVE MOMENT
A deployment choice should make sense to the people responsible for the service, not only the technical team. Connect the task, information and operating requirements first. Then choose the combination of local and hosted capabilities that fits, with a clear explanation of what the organisation is retaining and what it is delegating.
QUESTION FOR READERS
Could our process owner explain where a customer’s information goes during an AI-assisted task?
Brought to you by Data Wave — your AI & Data Team as a Subscription.
