Copilot Code: Microsoft Lets Non-Developers Build Internal Apps with AI

On September 25, 2026, Microsoft unveiled a deep overhaul of Copilot, now built around four pillars: Home, Code, Autopilot, and native Office integration. The piece drawing the most attention from the technical community is Code, which lets people with no development background describe an application, a dashboard, or an automation in natural language, then run it directly inside the company’s tenant.

#Four pillars for a unified Copilot

Home merges the Chat and Cowork experiences into a single entry point, with Word, Excel, and PowerPoint now embedded in the interface. Autopilot is a persistent agent handling recurring tasks, such as supplier follow-ups, without constant prompting. Code, for its part, targets tool creation by non-developers, running in sandboxed environments inside the organization’s tenant.

#Code: GitHub Copilot’s engine repurposed for no-code

According to Microsoft, Code runs on “the same underlying technology as GitHub Copilot,” combined with Microsoft IQ, the company’s enterprise intelligence layer. In practice, a natural-language description gets turned into executable code, packaged and hosted through the Microsoft Copilot Managed Runtime, the new execution infrastructure announced alongside it. That architecture echoes the manifests used by GitHub Copilot extensions, where a third-party service declares its capabilities in a structured format:

{
  "name": "internal-tool-agent",
  "runtime": "copilot-managed-runtime",
  "capabilities": ["generate_app", "run_sandboxed", "read_tenant_data"],
  "auth": { "type": "tenant_oauth", "scopes": ["Sites.Read.All"] }
}

This packaged-agent logic echoes how MCP already structures exchanges between models and tools on the developer side, but here Microsoft is aiming it at an audience that doesn’t code day to day.

#Governance: the question IT teams will ask first

This shift toward code generated by non-technical profiles connects to debates already documented around GitHub Copilot’s governance and billing in a professional context: who approves access to tenant data, who audits the code produced, and how to limit the spread of unmaintained automations. The topic also echoes practices already seen in running editorial agents in production, where human oversight remains the variable that determines how reliable the output is.

#A rollout spread over several months

Code runs in sandboxed environments inside the company’s tenant, on the same underlying technology as GitHub Copilot.

#Key takeaways

Microsoft isn’t launching a standalone product; it’s recomposing its entire Copilot lineup around app generation for non-technical users, building on GitHub Copilot’s already-proven infrastructure. The timeline stays gradual, limited to the Frontier program for now, giving technical teams time to prepare governance around data access and code auditing.

#Going further

Source: Microsoft, “Introducing the new Copilot with Home, Code and Autopilot”, September 25, 2026.