News
Microsoft’s Azure AI Foundry Is Quietly Building the Internet’s Brain
Microsoft is taking a big swing at the future of autonomous AI research—and it's doing it through Azure AI Foundry. The company just put OpenAI’s "deep research" model inside its developer platform, turning what was once a ChatGPT power-user tool into a composable backend engine for enterprise-scale reasoning.
Azure AI Foundry is now in limited preview with a major upgrade: agents that don’t just surf the web—they interrogate it. These aren't your average chatbots. Powered by OpenAI's o3-deep-research model, the agents are designed to perform structured, multi-step investigations, parse sources, analyze conflicting information, and return fully auditable reports with citations and reasoning trails.
"With Deep Research, developers can build agents that deeply plan, analyze, and synthesize information from across the web—automate complex research tasks, generate transparent, auditable outputs, and seamlessly compose multi-step workflows with other tools and agents in Azure AI Foundry," said Yina Arenas, VP of Product, Core AI at Microsoft, in a July 7 blog post.
From ChatGPT Feature to Programmable Research Stack
OpenAI first launched Deep Research in February 2025 as a research-focused capability inside ChatGPT. While most AI assistants aim for speed, Deep Research prioritized depth—designed for knowledge workers who care more about accuracy and auditability than glib responses. It draws from real-time Bing Search results, uses internal prompts to plan and revise its approach, and produces outputs that look less like chatbot blurbs and more like briefing documents.
Built on OpenAI’s o3 model, Deep Research can generate structured outputs that summarize, cite, and justify every insight. Microsoft tuned this for Azure by making it composable—so developers can call it via API, plug it into Logic Apps or Azure Functions, and link it with other agentic workflows.
"Unlike packaged chat assistants, Deep Research in Foundry Agent Service can evolve with your needs—ready for automation, extensibility, and integration with future internal data sources as we expand support," Arenas said.
Think Like a Researcher, Scale Like a Cloud App
Here’s what makes the Azure version different: it’s not a chat UI. It’s infrastructure. Developers can now build research agents into broader systems—like triggering a competitor analysis, synthesizing it into a presentation deck, and emailing it to execs—all as one multi-agent flow.
Microsoft calls this "reasoning automation," and it’s already making its way into products like Microsoft 365 Copilot. Inside apps like Word and Outlook, the Researcher and Analyst agents use Deep Research to gather intel from internal documents, chats, and emails, then turn it into executive summaries or data-driven insights.
"You can trigger a research agent as part of a multi-agent chain: one agent performs deep web analysis, another generates a slide deck with Azure Functions, while a third emails the result to decision makers with Azure Logic Apps."
Real AI Research, Not Just Search
What makes this system different from traditional LLM workflows is its agentic architecture. Deep Research doesn’t just generate answers—it plans how to find them.
A typical workflow starts with a scoping prompt, gathers Bing-grounded sources, and executes a step-by-step research strategy. The result is an output with internal prompts, source links, and a full reasoning trail—an audit log for how the answer was built.
Microsoft outlines several key features for Azure AI Foundry users:
- Automated, web-scale research with Bing Search grounding and OpenAI’s latest reasoning model.
- Composable agents that can be triggered by apps, other agents, or workflows.
- Workflow orchestration using Azure Logic Apps and Foundry Agent connectors.
- Enterprise-grade governance for compliance, access control, and observability.
Agentic AI That Actually Performs
Performance-wise, the o3-deep-research model isn’t just a demo. It hit 26.6 percent accuracy on Humanity's Last Exam, a massive benchmark spanning over 100 expert domains. It also topped charts on GAIA, a new standard for web-based reasoning, where it excelled at planning, synthesis, and evidence tracking.
The feature was originally adopted into Microsoft 365 to support use cases like business reporting, knowledge synthesis, and regulatory analysis. But in Azure AI Foundry, it levels up. Instead of being limited to user prompts, Deep Research becomes a programmable intelligence layer for apps and services.
Pricing for the o3-deep-research model starts at $10 per million input tokens and $40 per million output tokens, with discounted rates for cache hits. GPT-based intent scoping and Bing Search are billed separately. The preview is available to approved customers via limited access for now.
What’s Next: From Research to Real-World Action
Microsoft and OpenAI see Deep Research as just the start. OpenAI has hinted that it will eventually combine research agents with execution agents like Operator, which can not only analyze but also take action—like using a browser to complete a task for the user.
In that vision, agents don’t just answer questions. They make decisions, trigger processes, and move systems forward.
Azure AI Foundry’s latest update is a clear signal: the era of agentic AI isn’t theoretical anymore. It’s here—and it’s programmable.
About the Author
David Ramel is an editor and writer at Converge 360.