Microsoft releases AI agent platforms for research and development

Microsoft announced that its Discovery platform is now generally available, offering a production‑ready environment that lets scientists and research teams use AI agents to coordinate data analysis, hypothesis generation, experimentation and knowledge management. The service, unveiled at Build 2026, includes a graph‑based knowledge engine that links proprietary research data with external scientific information, and a preview desktop app that can be downloaded from GitHub using a Copilot account.

At the same event Microsoft also moved its Agent Framework to version 1.0 GA, introducing the Agent Harness and Foundry Hosted Agents as a supported runtime for building, running and governing multi‑step AI agents. The harness provides built‑in capabilities such as function invocation, history persistence, tool approval and OpenTelemetry, while the hosted service bills on consumption. Principal software engineer Wes Steyn described the harness as the core that enables models to plan, execute and iterate on tasks.

Both releases aim to embed agentic AI across Microsoft’s portfolio and give enterprise and academic users a governed, reproducible way to incorporate AI into R&D workflows.