AI Agents in PowerShell: Skills Over Prompts

The last stream showed that AI agents are simple.
This one shows how they stay simple as they grow.
Claude Skills helped popularize the idea of skills-based agents. That implementation is bash-centric.

This session explores the same architectural pattern, implemented natively in PowerShell.

Skills aren’t a Claude feature.
They’re a design pattern.

Bash is one implementation. PowerShell is another.

In this livestream, we evolve a basic PowerShell AI agent into a reusable system by replacing prompt sprawl with structured, discoverable skills.

Skills are defined as folders with metadata and intent, not giant prompts stuffed into a context window.

We’ll cover:

• Why skills beat prompts as agents grow
• Implementing Claude-style skills in native PowerShell
• Skill discovery and progressive disclosure
• Safety rails and constrained execution
• Keeping agents composable without frameworks
• Why PowerShell’s object model matters here

This session is for builders who already got an agent working and immediately asked, “How do I not ruin this?”

About Doug
Doug Finke is a 16-time Microsoft MVP and creator of the PowerShell AI module — chat in your console, NLP in your scripts — plus other popular tools like PowerShell + Excel integration.

With 40+ years in tech (from punch cards to AI), Doug now focuses on the next evolution: building autonomous tools and showing others how to think, type, get with AI — not just code around it.