I'm trying to automate our project scheduling. I need to understand how Microsoft Semantic Kernel powers enterprise AI when the steps aren't linear. If a resource is unavailable, can the planner pivot and suggest a different timeline automatically using our internal PM tools?
3 answers
The Planner is essentially the "brain" that looks at your available "Plugins" and the user's goal to create a roadmap. In your case, you would create a "ResourcePlugin" and a "SchedulingPlugin." When the user asks for a timeline, the Planner sees that it needs to check resources first. If the resource check returns "unavailable," the LLM can use that feedback to loop back and try a different date. This iterative reasoning, often called "Chain of Thought," is exactly how Microsoft Semantic Kernel powers enterprise AI to handle non-linear, real-world business scenarios.
Is there a way to visualize the plan before it executes? I’m worried about the AI making a mistake and sending out a wrong schedule to the whole department.
We use the Function Stepwise Planner for this. It’s slower but much more accurate for complex tasks that require multiple API calls to different PM tools.
We found the same thing, Ryan. The stepwise approach is better because it 'thinks' after each step, which is crucial when dealing with changing project data.
Jeffrey, you can definitely do that. The Planner returns a "Plan" object (often in JSON or XML format) before execution. You can display this in your UI for a "Human-in-the-loop" approval. Once the manager clicks 'Approve,' the Kernel then proceeds to execute the steps. This safeguard is a best practice for any enterprise AI implementation to ensure accuracy and human oversight.