Agents, Agents and Agents
I am a Software Engineer at Microsoft, where I co-engineer Agentic solutions alongside Azure’s largest global customers. I don’t just talk about AI; I’m in the trenches solving the messy, real-world problems that come with production-grade agents. My blog is a look behind the curtain—from the friction of aligning with business experts to the technical hurdles of implementation. If you’re looking for honest insights on building AI that actually works (and the lessons learned when it doesn't), you’re in the right place.
Markdown supported · ⌘↵ to post
klement Gunndu
The observation about agents optimizing locally instead of for system-level success is exactly what we hit in production. We run multi-agent orchestration where each agent has bounded autonomy and a shared state machine — but the coordination still breaks when agents interpret ownership differently. Cursor's pattern of structured task state over natural language handoffs matches our finding: deterministic routing between agents with LLM judgment only at decision points, not coordination.