Microsoft Azure Publishes a Guide to Help Teams Pick Between AI Skills and Sub-Agents

Microsoft Azure has released new architecture guidance to help developers decide between two growing ways to extend AI systems: AI Skills and Sub-Agents. As companies add AI into daily workflows, engineers often face the same question: should they teach the AI a new ability, or should they hand the job to a smaller helper model? Azure's new guide tries to give a clearer answer.

AI Skills are best understood as "manuals" for a language model. They sit at the knowledge layer and describe how a task should be done in plain language. When a request matches, the model loads the skill only for that moment. This keeps the context window light and is well suited to repeated work such as reviewing code, summarising reports, or following company style rules.

Sub-Agents work at a different level. They are small, specialised models that can plan, call tools, and finish a piece of work on their own, before reporting back to the main agent. This makes them stronger for complex, multi-step jobs, but they bring more overhead in cost, latency, and debugging. Azure's guidance recommends Sub-Agents when one task clearly stands apart from the rest of the workflow.

The bigger message is that the AI tooling market is splitting, not converging. Choosing between Skills and Sub-Agents is becoming a routine architecture decision, much like picking a database or a message queue. Teams that learn to mix both will move faster, while teams that pick the wrong layer for the job often pay the price in slow responses and rising bills.