Agents for us all
Microsoft’s Azure AI Foundry hard launched this week
AI agents aren’t new. But many are stuck in basic mode, answering simple questions or following rigid scripts. Microsoft’s Azure AI Foundry, launched late 2024 and publically upgraded this week at Build 2025, changes the game. It builds agents that take action—fixing code, streamlining shipping, digging into medical data—with real independence. How does it compare to other AI platforms? What are developers building? Let’s cut to the chase.
Foundry: A Toolkit for Smart Agents
Foundry is a platform for creating AI agents: programs that think, act, and work together on tasks. It offers over 1,900 AI models (like xAI’s Grok 3 or OpenAI’s GPT-4.5), tools to coordinate agents, and security to keep things tight. Developers can build agents quickly, deploy them to tools like GitHub or Microsoft Teams, and track performance with real-time data. Fujitsu used it to cut sales proposal times by 67%. That’s real impact, not buzz.
Foundry runs on Microsoft’s cloud, with GitHub Copilot and Visual Studio Code as allies. It uses a system to manage multiple agents (like a conductor for a band), letting one handle data while another triggers actions. Safety tools test for errors or biases, security IDs lock down agents, and performance tracking catches slowdowns. For devs, it’s straightforward. You get integration with Visual Studio Code (a popular coding editor), options to tweak models, and dashboards to spot issues like slow responses.
Why It’s a Big Deal
Foundry backing open standards for AI agents isn’t some minor product update… It’s a bet on how the next wave of compute will be coordinated.
Agents don’t work well inside walled gardens. They need to move fast, talk across platforms, and make decisions on the fly. By going open, Microsoft positions itself as the backbone of this new infrastructure, while also sidestepping the kind of geopolitical chokepoints we’re starting to see in the U.S.–China AI race.
This shift is already showing up in the real world. C.H. Robinson’s shipping agents have automated millions of tasks like pricing and order routing. Merck’s researchers are using agents to find drug candidates faster than ever. And small businesses are quietly using agents to manage inventory, respond to customers, and handle ops at scale. From an economic perspective, this means soon (and already) collapsing the cost of coordination, across every sector.
But there’s an even bigger thing. As agents become critical to how companies compete, the fight won’t just be about AI models, but who owns the rails they run on. Europe wants to regulate it. China wants to centralize it. The U.S. is inching toward modular, open coordination. Foundry isn’t leading the agent market—but it may end up defining the ground rules everyone else has to play by. (This is due its very own deep-dive, coming out end of month!)
Devs in Action: What Agents Do
Agents are live and delivering. Here’s what developers are making:
Customer Service: Cineplex’s agents handle refunds in 30 seconds. They pull data from customer systems and act—no human needed.
Business Insights: NTT DATA’s agents analyze HR records using Microsoft’s data platform, spotting trends fast. It’s like a super-smart spreadsheet.
Code: GitHub Copilot’s agents review code for 15M+ developers. They catch bugs and speed up releases.
Healthcare: Stanford’s agents prep cancer treatment plans. Others scan medical journals, speeding drug discovery.
Logistics: C.H. Robinson’s agents optimize shipping routes. Hershey’s agents suggest recipes based on what you buy.
Foundry vs. Other Platforms
How does Foundry stack up against Google Vertex AI, AWS Bedrock, and TensorFlow (a library for building AI models)?
Azure AI Foundry:
Wins: Tight integration with Microsoft’s ecosystem (GitHub, Teams, Azure). Built-in tools for multi-agent coordination, model monitoring, and safety. Strong support for enterprise workflows.
Loses: Heavy Microsoft emphasis; less flexible outside that ecosystem. Some models and features are geo-restricted (mostly US).For: Devs building secure, team-based agents for firms like KPMG or Fujitsu.
Wins: Polished tools for training, deploying, and managing models—especially for structured data. Easy AI for beginners with tools for data analysis. Strong AutoML and integration with Google Cloud’s data stack.
Loses: Less native support for agent orchestration. (So far! Not for long…) More tuned for ML pipelines than autonomous workflows.
For: Data scientists and analysts focused on model development, not agent coordination.
Wins: No setup needed. Offers models from Anthropic + more. Quick to start.
Loses: Lacks deep coding-tool integration. Agent teamwork lags.
For: Fast prototyping of simple AI apps, like retail bots.
Wins: Full control. Open-source, portable, and backed by years of community support. Great for custom model architecture.
Loses: No built-in agent framework or enterprise controls. Scaling and deployment require manual setup.
For: Researchers experimenting with AI or engineers building custom architecture, not production systems.
Visual Comparison:
Take: Foundry’s for devs building agent teams for business. Vertex AI’s for data analysis. Bedrock’s for quick apps. TensorFlow’s for research. X agrees: “Foundry’s agent coordination is next-level.” Over 70,000 customers are on board.
The Risks: Not Magic, Not Yet
Agents can still get things wrong—bad outputs, missed context, security gaps. Foundry has strong safety tools, with runtime monitoring and stress testing built in. But it’s not foolproof. Some of the most capable models are US-only, and the tight Microsoft integration can feel limiting if you're not fully in that ecosystem. Sarah Bird said it well: tests need to outsmart risks. That’s the bar.
What’s Next: Agents as Infrastructure
Build 2025 quietly signaled a shift. Foundry now supports images, audio, local on-device AI, and scaled coordination across huge numbers of agents. This isn’t about flashy demos—it’s about turning agents into core infrastructure. Teams will tweak models to clean clinical data, scan financials, write software, run research. Not as side tools, but as primary execution layers. One internal line captured it perfectly: agents are the new microservices.
Your Turn
Foundry’s raising the bar. How’s it compare to your needs? Devs, what agents are you building? Tech leads, what’s the next big use case? Email me at ciphertalkpodcast@gmail.com or comment. Let’s talk.
Stay sharp,
/m