Foundry’s Agentic Bet: From Single Prompts to Production AI Agents
Microsoft Foundry is an end-to-end AI agent infrastructure platform that combines frontier models, production AI agents, and governance controls so organizations can move from one-off prompts to reliable, multi-step agent workflows that live inside real business systems and tools. Microsoft is no longer selling Foundry as an optional sidecar to Azure; it is positioning it as the default home for Microsoft Foundry agents, from development to deployment. More than 100,000 organizations already build on Foundry, and names like Adobe, Telefónica, and Tata Consultancy Services are running agents in production. That scale matters: this is not a proof-of-concept platform, it is a bet that the agentic era software stack will be centralized in one place. Instead of stitching together separate services, Foundry promises a single pipeline to build, run, and distribute production AI agents.
Frontier Models Deployment: GPT-5.6 as the Agent Workhorse
The real engine behind Microsoft Foundry agents is the new GPT‑5.6 frontier model series, now generally available in Foundry Models and Foundry Agent Service. GPT‑5.6 Sol targets extended reasoning, agentic workflows, and code-focused workloads for demanding enterprise scenarios. Terra offers a balanced option for everyday work at lower cost than GPT‑5.5, while Luna is optimized for high-volume, low-latency use. According to Microsoft, “More than 100,000 organizations are already building on Microsoft Foundry, and companies like Adobe, Telefónica, and Tata Consultancy Services are running agents in production today.” Crucially, these models are available across 28 global regions and multiple Data Zones from day one, so teams can adopt frontier models where their applications already run. This broad frontier models deployment turns GPT‑5.6 into a practical workhorse rather than an isolated research artifact.
| Model | Role in Agents | Pricing (USD/MYR) |
|---|---|---|
| GPT‑5.6 Sol | Extended reasoning and agentic workflows | Input USD 5.00 (approx. RM23.00), Output USD 30.00 (approx. RM138.00) |
| GPT‑5.6 Terra | Balanced everyday workloads | Input USD 2.50 (approx. RM11.50), Output USD 15.00 (approx. RM69.00) |
| GPT‑5.6 Luna | High-volume, latency-sensitive tasks | Input USD 1.00 (approx. RM4.60), Output USD 6.00 (approx. RM27.60) |
Hosted Agents and Governance: The Missing Production Layer
The most important change is not the models but the runtime: hosted agents in Foundry Agent Service are now generally available, giving developers a single production environment for agents built with Microsoft Agent Framework, GitHub Copilot SDK, and other harnesses. Foundry’s AI agent infrastructure wraps those agents in enterprise identity, security, and compliance, including network isolation through Azure Virtual Network integration. Toolboxes connect agents to systems and data, while upcoming memory and routines in public preview hint at more stateful, long‑running behavior. Publishing to Microsoft 365 Copilot and Microsoft Teams becomes generally available next week, meaning agents move directly into the applications where hundreds of millions of people work. This is the production layer many organizations lacked: instead of fragile scripts, they get observable, governed agent services shaped to long‑term business value.
Development Experience: Agent Framework, Go, and Multi-Agent Workflows
On the build side, Microsoft is aggressively compressing developer friction. Agent development starts where engineers already work—in GitHub Copilot and VS Code—with the Foundry Toolkit handling deployment. Microsoft unified AutoGen and Semantic Kernel into a single Agent Framework, introduced in October 2025 and now at general availability, to give one supported way to construct AI agents and multi-agent systems. Recently, it brought Microsoft Agent Framework to Go in public preview, giving Go developers first‑party support for models from Microsoft Foundry alongside Azure OpenAI and other providers. Until now, Go teams had to juggle raw HTTP calls, sidecar Python or Node processes, or uneven community libraries to embed AI agents. With native Go support, cloud‑native services, background workers, and command‑line tools can coordinate multi‑step agent workflows directly, which fits the infrastructure‑heavy world Go already dominates.
Why This Shift Matters for the Agentic Era
Foundry’s evolution shows a clear conviction: the agentic era will be won by platforms that make agents a first‑class production concern, not side projects. Microsoft promises that developers can build agents where they already work, run them on infrastructure they trust, and put them in front of users without stitching disconnected platforms. With Global, Data Zone, and Regional deployments, organizations align frontier AI with sovereignty, compliance, and performance while keeping a consistent development flow. The new Asia‑Pacific Data Zone follows this logic, allowing customers to run frontier OpenAI models while keeping processing in‑region. In practice, this means AI agents can stop being prototypes and become standard parts of enterprise software. If Microsoft can keep Foundry open across models and frameworks while deepening governance, it will be well positioned as agentic era software leaves the hype cycle and joins everyday operations.
