Why Microsoft is done chasing the AI model of the month
Content
Within the past few days and weeks, we have seen AI models announced from Grok 4.7, Claude Opus 5.5 on Monday, and GPT-6 Sol and Luna on the same day. Just a few hours after, Sonnet 5.5, GPT-6.1 Sol and Google‘s latest frontier model Gemini 4 Argon. Since the release of ChatGPT in late November 2022, close to 90 new models were announced in total by the top contributors, and more than 1000 in general.
However, models change much faster than business processes.
Now I ask the important question: how many enterprises rebuild their core process because a model benchmark moved? Very few would even consider it. Customer data, approvals, integrations, permissions and regulatory requirements do not disappear when a stronger model ships.
Of course, the model matters; it changes quality, latency and cost. But it should remain replaceable, and if changing the model means rebuilding the business process, the architecture is wrong.
Microsoft is building models, but selling the platform
Microsoft is not absent from model development. In June 2026, Microsoft AI announced seven in-house MAI models spanning reasoning, coding, image generation, transcription and voice. The same announcement says the models will be optimized for Microsoft products and distributed through Microsoft Foundry and external developer platforms.
My conclusion is different from Microsoft’s product wording: the company does not need to win the public conversation with a new general-purpose frontier model every month. Its stronger enterprise position is the layer around the model. Let’s examine it:
Copilot Home: one front door for work
Microsoft announced Home as the new starting point in Copilot, bringing Chat and Cowork together and putting Word, Excel and PowerPoint into the Copilot experience. Microsoft also says a future routing experience will direct work to Chat, Cowork or Code based on the request.
What does this mean? Well, it means that Microsoft wants Copilot to become the entry point to work, while the platform chooses the right execution mode behind it.
Copilot Code: citizen development returns
Microsoft says Code can create an app, tracker, dashboard, automation or workflow from a natural-language description. It uses the same underlying technology as GitHub Copilot, runs in a sandboxed environment and can be hosted securely inside the customer tenant.
This is the Power Apps idea moved forward. More people can create software, while IT still needs control over identity, data, deployment and operations, and the authoring method changed from low-code composition to intent-driven code generation.
Autopilot: persistent work, not another chat window
Microsoft describes Autopilot, previously called Scout, as a persistent, proactive and personal agent. It can watch channels, follow up on threads, run recurring work and resume projects without a new prompt. Microsoft also says it lives in the tenant with its own identity, memory, computer and workspace, backed by permissions, audit and governance.
Work IQ: reusable business context
Microsoft says Work IQ grounds Copilot and agents in business data from Dynamics 365 and Power Platform. The preview started on September 30, 2026, with rollout continuing through October. The announcement covers semantic models, reusable business skills and governed actions that can update source systems within the user’s permissions and required approvals.
This is the strongest evidence for the platform thesis: A model can reason over language, but doesn’t automatically know a company’s inner workings (e. g. approval processes or business vocabulary). Work IQ is Microsoft’s attempt to make that context reusable across Copilot and agent experiences.
Managed Runtime: build where you want, operate under one control model
Microsoft introduced Copilot Managed Runtime in public preview as an enterprise-grade platform for running code inside the Microsoft 365 tenant boundary. It provides Microsoft-operated hosting, Entra identity, governed data access, lifecycle management, Git-based version and source control, and a central inventory in the Microsoft 365 admin center. Microsoft states that it already powers apps created in Cowork, Code and Copilot Studio and is being opened to compatible third-party tooling and professional developers.
Microsoft quotes Lovable’s partnership lead. saying an app built with Lovable can run in the Microsoft tenant with the same sign-in, policies and app inventory. Crazy times.
Plugin registry: tools become governed platform assets
Microsoft announced a unified registry for Microsoft, partner and custom-built plugins. IT can approve and manage plugins centrally, while developers and partners can publish once for supported Copilot experiences.
That matters because enterprise agents need controlled access to tools and actions. A separate connector and permission model for every agent would only create a new integration problem.
Auto and FinOps: model selection and cost control
Microsoft says Auto evaluates accuracy, speed and cost to route each request to a suitable model under the user subscription license. For agentic workloads, usage-based billing gives organizations explicit model choice and cost control. The announcement also adds spending policies, model-family controls, credit approvals and usage visibility under FinOps for AI.
The practical message is blunt: most employees should not spend their day selecting models. The platform should route ordinary work, while specialist teams retain control where quality, regulation or economics require it.
Is Salesforce making the same bet?
Salesforce and Anthropic announced Claudeforce in August 2026. The launch combines Claude with Salesforce data, workflows, business logic, actions and governance. Salesforce in Claude starts with a plugin containing 37 prebuilt sales skills. The announcement also says Claude is available inside Salesforce products and is the default model across several Salesforce and Slack experiences.
The architectural parallel is clear: probabilistic reasoning comes from the model; enterprise execution depends on data, rules, permissions and workflows. The open question is a strategic one: Is Anthropic one model partner inside a model-diverse Salesforce platform, or does Claude become the interface that owns the user relationship?
Microsoft is returning to what it knows best
With Microsoft’s announcements, a new path emerges: Home becomes the new entry point for all users, with access to Copilot Chat and Cowork. For application creation and code work, Microsoft adds GitHub Copilot features to the very same Copilot App, while Autopilots complete the tasks in the background.
All of these features are grounded in Work IQ, an unique business context supplier for your own preferences and business focus. All supported by Managed Runtime, that allows you securely host any app you are going to develop within the same framework. Everything in one place, governed and controlled in the new FinOps economics insights.
Home becomes the entry point. Code broadens application creation. Autopilot runs persistent work. Work IQ supplies business context and governed actions. Managed Runtime hosts and operates generated software. The plugin registry distributes approved capabilities. FinOps manages the economics.
That is more than a collection of Copilot features. It is an enterprise business platform.
The pattern resembles what made Power Platform successful: broad creation at the edge, central control underneath, and deep connections into productivity software, business applications, identity and data. Microsoft is now applying that playbook to AI.
MAI still matters. It gives Microsoft its own models, product optimization and negotiating leverage. But Microsoft does not need to win every monthly benchmark to win the enterprise market. It needs to make the model replaceable.
Model selection is not the holy grail. Processes, data, identity, security, orchestration, governance and operations decide whether an AI transformation survives beyond the pilot.
The next model will arrive anyway. The real test is whether your business process has to be rebuilt when it does. Microsoft is building toward an answer: it should not.
RECENT POST
OUR EVENTS
