What happened
Factory’s November 18, 2026 announcement expands its collaboration with Microsoft. Factory says organizations can deploy its software development agents, called Droids, within their Azure environments and existing identity, network and source-control systems. Droids can run in customer-controlled Azure Virtual Network (VNet) environments. Factory’s self-hosted orchestrator for agent sandboxes and provisioning backend, OpenCompute, keeps agent workloads in the customer environment, on its Azure Kubernetes Service (AKS) clusters or VNet. Droids can use models served through Microsoft Foundry, which Factory says gives enterprises the model choice and governance available in their Azure environments.
The workflow is designed to take engineering items from initial report to a proposed or merged fix. Organizations can hand off issues from Jira, Linear or GitHub, or route production alerts to Factory. Droids triage items, investigate root causes, write and test fixes, then open pull requests for human approval. Factory says customers can also configure automated final merges. In GitHub or Azure DevOps, Droids can open pull requests, respond to comments and act on CI results within existing review rules and approvals. For security work, Droids can review GitHub pull requests, trace changed data flows, and post validated findings with severity and suggested fixes. Scheduled full-codebase audits are also supported; Factory says a Droid can take an audit finding through to a tested fix.
The announcement also covers execution controls on developers’ machines. With Microsoft Execution Containers (MXC), every command a Droid runs executes inside a container that can access only files and network destinations approved by the team. Factory says the operating system enforces those limits and IT teams manage policies through Microsoft Intune. In Microsoft Teams, users can mention @Factory in standard or private channels, group chats or one-to-one chats to hand off tasks; Factory says Droids can use earlier messages as context.
Factory describes a broader “self-improving software factory” model: engineers set goals and govern outcomes, while Droids execute across the software development lifecycle. Signals from code review, testing and CI, production incidents and delivery analytics feed back into Factory. Factory says the service is available today in Microsoft Marketplace and that eligible purchases count toward a customer’s Microsoft Azure Consumption Commitment (MACC). It directs interested customers to their Microsoft account team or Factory sales.
What the archive adds
New Runtime covered Factory and Databricks on enterprise modernization as one engineering workflow on October 22, 2026. This announcement is a new event, not a restatement: it adds deployment and execution details for Azure, including network-contained workloads, MXC command controls and Teams handoffs.
Trend signal
This supports agent security moving to runtime boundaries: Factory describes operating-system-enforced limits on the files and network destinations an agent command can reach, with policy administration through Intune. The claim is a vendor description, not an independent security evaluation, but the design places controls in execution rather than relying only on instructions to the agent.
Why now
Factory’s announcement frames the expanded collaboration around enterprises seeking to apply agents to software development while retaining control over code and governance. It also places the announcement alongside existing Azure, GitHub, Teams and Microsoft procurement workflows, rather than presenting a separate standalone environment.
Why it matters
For teams considering coding agents in environments with strict controls, the deployment boundary is part of the product decision. Running the orchestrator and workloads within an Azure VNet or AKS cluster may align agent execution with existing network and identity arrangements; MXC adds an explicit restriction on which files and destinations commands can access. These are relevant controls, but they do not by themselves establish that a deployment meets any particular organization’s security requirements.
The described workflow also keeps a recognizable review path: an agent can investigate and test a fix, while a pull request remains subject to existing rules and human approval. Teams can evaluate automation without immediately handing over merge authority, and can reserve automated final merges for customer-defined preferences. For adoption, the useful question is not simply whether an agent can write code, but whether its access, evidence and handoff fit the team’s operating controls.
Limits and uncertainty
Factory’s announcement does not quantify deployment performance, fix quality, security outcomes, or the time and cost required to operate this setup. It does not provide an independent assessment of MXC enforcement or describe the specific policy configurations customers should use. “Self-improving” is described through feedback signals from reviews, tests, incidents and analytics, but the announcement does not explain the update mechanism or how customers inspect and control changes. Availability is stated for Microsoft Marketplace, but no prices or deployment prerequisites are given.
Reader takeaway
Factory is pitching an Azure-resident coding-agent workflow with customer-controlled execution boundaries and existing software-review gates. Teams evaluating it should distinguish these concrete deployment controls from broader claims about autonomous or self-improving engineering, then validate the actual access policies and approval behavior in their own environment.