A laptop displays a cybersecurity dashboard linking cloud services, code repositories, and access controls.
Factory's Azure-hosted coding agents now have a Microsoft containment layer on the developer's PC, but plenty of detail is still missing.

Factory announced on November 18, 2026 that it is expanding its collaboration with Microsoft to give enterprises greater flexibility to deploy self-improving software factories. The pitch is that autonomous coding agents can run inside the Azure, GitHub and Teams estate that many enterprises already govern. The announcement touches Azure, Foundry, GitHub, Azure DevOps, Teams, Intune and a new Windows containment technology. That is a lot of Microsoft surface area for a single press release.

This piece separates what Factory claims from what Microsoft's own documentation supports, and lists what a cautious IT team should still verify.

A laptop displays a cybersecurity dashboard linking cloud services, code repositories, and access controls. What Factory says it is delivering​

Factory is the vendor here, so every product claim below is the company's own.

Azure deployment. Organizations can run Factory within their Microsoft Azure environment, using existing code, identity, and governance systems. Factory calls its agents "Droids". They run in Azure, including customer-controlled Azure Virtual Network (VNet) environments. Factory also describes a self-hosted orchestrator for agent sandboxes and a provisioning backend called OpenCompute. It says these keep workloads on the customer's own Azure Kubernetes Service clusters or VNet.

Foundry models. This allows Droids to use models served through Microsoft Foundry, giving enterprises the same model choice and governance already available in their Azure environments. The release doesn't say which Foundry models are supported.

Source control and CI. Droids can then open pull requests, respond to comments, and act on CI results in GitHub or Azure DevOps, within existing review rules and approvals.

Intake and resolution. Work can arrive from Jira, Linear or GitHub, or as routed production alerts. Factory says Droids triage each item, find the root cause, write and test a fix, and open a pull request for human approval. They can also merge automatically if the customer's preferences allow it.

Security review. Droids review pull requests in GitHub for security issues, tracing changed data flows and posting validated findings with severity and suggested fixes. Organizations can also run scheduled full-codebase audits, and Droids can carry each finding through to a tested fix. The release gives no detection-accuracy figures and doesn't explain what "validated" means. Treat that word as a claim to test, not a result.

Teams. Users can mention @Factory in standard and private channels, group chats and one-on-one chats. Factory says the agent uses earlier messages as context. Teams admins will want to know what permissions and consent the app needs, which message content leaves the tenant, and how it is retained. The announcement says nothing on any of these.

The MXC piece: the most interesting part for Windows admins​

Factory says its protection now extends to the developer's machine. Every command a Droid runs executes in a container limited to team-approved files and network destinations. Factory says the operating system enforces those limits and that IT manages the policies through Microsoft Intune.

Microsoft's own description of Microsoft Execution Containers (MXC) gives this some grounding. It is a policy-driven execution layer for untrusted code or dynamically generated workloads. Microsoft's blog says MXC is now generally available. Its Learn documentation describes the technology as open source, with .NET, TypeScript and Rust SDKs.

"Container" is not one thing​

The Windows Developer Blog post from October 7, 2026 lists several backends:

BackendWhere it runsNotes
Process containerWindows 11, macOS, LinuxLightweight. Uses AppContainer on Windows, Seatbelt on macOS and Bubblewrap on Linux.
Session containerWindows 11 onlySeparate Windows account and session, with its own desktop, clipboard and input boundaries.
WSL containerWindows 11 onlyLinux-first toolchains through WSL.
MicroVMWindows 11 and Linux, experimentalHardware-backed isolation for higher-risk workloads.

Microsoft states that each containment backend has distinct security properties and that workloads should be evaluated for fit. Its Learn page makes a similar point: available restrictions depend on the MXC release, the backend and the host. So "Factory uses MXC" does not tell you how strong the isolation is. Factory's release doesn't say which backend it uses.

What containment does not do​

Microsoft's documentation is blunt. Containment limits what a workload can do, but it doesn't make generated instructions, outputs or permitted actions trustworthy. Microsoft says to keep policy decisions in trusted host code and validate results before acting on them.

In practice, an MXC boundary can stop an agent process from reading your Documents folder or phoning an unapproved host. It cannot tell you whether the patch is correct. It also can't tell you whether an approved network destination is safe, or whether an auto-merge should have happened. Factory's feature set includes optional automatic merges, so branch protection, required reviewers and CI gates remain the controls that matter most.

The Intune timing question​

This is the most important open item. Microsoft's October 7 post said Intune policy for managing MXC process containers used by MXC-integrated agents on Windows 11 will soon be available. Factory's November 18 release states in the present tense that IT teams manage its policies through Intune.

The sources don't reconcile the two. They don't show whether Microsoft's Intune capability has shipped since October. They also don't show what Factory's integration relies on, or which Windows and agent versions are required. Admins should confirm the current Intune policy surface with Microsoft and Factory before building a rollout plan around it. Public Microsoft material from Build coverage described MXC integration with Intune, Entra, Defender and Purview through Agent 365. That coverage described a roadmap, not a Factory-specific feature.

Microsoft also says MXC process containers on Windows can produce activity reports, with enforcement, learning and permissive modes. These are useful when drafting a least-privilege policy for a new agent. The sources don't say whether Factory exposes them.

Procurement: not entirely new​

Factory says it is available today in Microsoft Marketplace, and eligible purchases count toward a customer's Microsoft Azure Consumption Commitment (MACC). That is useful for buyers who want to draw down an existing commitment. It isn't a first, though. Factory announced Azure Marketplace availability with MACC eligibility back in July 2025. The new release extends that relationship; it doesn't introduce Marketplace or MACC eligibility.

Factory's older Marketplace post also cites figures such as 2–4× faster feature development. Those are vendor listing claims, not independent measurements, and this announcement doesn't supply new evidence for them.

What the endorsements prove​

Factory CEO Matan Grinberg says regulated enterprises need autonomy without giving up control of their code. GitHub COO Kyle Daigle calls the collaboration a governed path to deploy agents across the software development lifecycle on Microsoft Azure. These are stakeholder statements in a company press release. They show Microsoft is comfortable being associated with the product. They are not security validation.

What IT and security teams should ask before piloting​

  1. Which MXC backend does Factory use on Windows endpoints? Process, session, WSL container or MicroVM isolation make different guarantees.
  2. What does the default policy allow? Check file paths, credential locations and network destinations. Test both allowed and denied operations, as Microsoft's documentation advises.
  3. Is Intune management live for your tenant and Windows build? Get the specific policy names and prerequisites in writing.
  4. What can the Azure-side agent do with its identity? Review service principals, repository permissions and Foundry access under least privilege.
  5. Is automatic merge disabled by default? Keep required human review on protected branches until you have evidence about task quality.
  6. What leaves the tenant from Teams? Check admin consent, data flows and retention.
  7. Which Foundry models are supported, and under which governance settings?

Bottom line​

For organizations standardized on Azure, GitHub and Windows 11, this is a credible and well-timed integration. It combines a vendor's agent platform with Microsoft's new OS-level containment story. The caveats are real. The public material doesn't document version requirements, deployment steps, independent security testing or Factory-specific Intune behavior. It also doesn't settle the gap between Microsoft's "soon" in October and Factory's present-tense claim in November. Treat the announcement as a reason to run a scoped pilot with a strict policy. It is not proof that autonomous coding agents are now safe to leave unattended.

 

References

  1. Factory Expands Collaboration with Microsoft to Bring Autonomous Software Engineering to Enterprises - 01net 01net 2026-11-18T22:04:00+00:00
  2. Microsoft Execution Containers: Policy-driven containment for AI agents - Windows Developer Blog blogs.windows.com
  3. Microsoft Execution Containers: Policy-driven containment for AI agents - Windows Developer Blog blogs.windows.com