Agent Skills | LIVE150
Addy Osmani joins Burke Holland to break down what “agent skills” are, how they encode senior engineering judgment into repeatable workflows, and how to apply them to get more consistent results from AI coding tools.
Overview
What “agent skills” are
- Agent skills are structured workflows that capture experienced engineering judgment so an AI agent can follow a consistent process.
- The goal is to reduce variance in outcomes and make results more repeatable and closer to production quality.
Structuring skills around the SDLC
The session organizes skills around the software development life cycle:
- Define
- Plan
- Build
- Verify
- Review
- Ship
A design pattern for writing a skill
Addy describes a repeatable structure for each skill, including:
- Overview
- Rationale
- Red flags
- Verification steps
Example project discussion: habit tracker
- The speakers introduce a habit tracker project idea.
- They discuss defining MVP success and use GitHub-inspired analogies to frame the work.
Verification and review workflows
- Discussion on balancing test-driven workflows with review-based workflows.
- Emphasis on simplifying code and iterating with testing (including TDD).
Security and performance review considerations
- Security and performance review is discussed in the context of “vanilla JavaScript”.
- The session also touches on trust and security concerns when reviews are nondeterministic.
Resources
Speakers
- Addy Osmani
- Burke Holland
Event context
- Microsoft Build 2026 (session LIVE150)