Uploaded June 2026 | Updated September 2026, 2 weeks ago
Skills are quickly becoming the new unit of software, but most of them are written once, never versioned, and never tested. In this episode, Oleg from the Docker dev team sits down with a guest from Tessl to talk about how AI-native development is changing the way we build software. They dig into why skills need real package management, how evals prove whether a skill actually adds value, and how Docker sandboxes keep AI agents isolated and safe. Along the way they cover practical cost savings, the integration between Tessl and Docker sandboxes, and why contributing skills to GitHub without versioning or quality control can do more harm than good.
What you will learn:
• Why skills are the new unit of software, and how they compare to classes, packages, and npm modules
• How skills get assembled into larger reusable units like plugins and extensions
• Why AI context "rots" the same way software does, and why versioning and a registry matter
• What evals are, and how comparing model behavior with and without context measures real "lift"
• How a lack of lift can signal a skill is now obsolete, saving you tokens and money
• Why a cheaper model with good context can outperform an expensive model without it
• How the Tessl registry adds quality, security, and eval scores to skills
• How Docker sandboxes isolate AI agents using micro VMs, secrets injection, and network rules
• How Tessl "kits" provision tools and skills inside a sandbox for repeatable, shareable setups
🔥 Want More Docker Content?
If you found this demo exciting, hit that like button and subscribe for more! We’ve got even more Docker demos coming your way in this ongoing series showcasing new tools, integrations, and powerful workflows to level up your projects. Stay tuned!
Where to find Docker:
Docker: docker.com
LinkedIn: linkedin.com/company/docker
Bluesky: https://bsky.app/profile/docker.com
X: @docker
Instagram: @dockerinc
Skills are quickly becoming the new unit of software, but most of them are written once, never versioned, and never tested. In this episode, Oleg from the Docker dev team sits down with a guest from Tessl to talk about how AI-native development is changing the way we build software. They dig into why skills need real package management, how evals prove whether a skill actually adds value, and how Docker sandboxes keep AI agents isolated and safe. Along the way they cover practical cost savings, the integration between Tessl and Docker sandboxes, and why contributing skills to GitHub without versioning or quality control can do more harm than good.
What you will learn:
• Why skills are the new unit of software, and how they compare to classes, packages, and npm modules
• How skills get assembled into larger reusable units like plugins and extensions
• Why AI context "rots" the same way software does, and why versioning and a registry matter
• What evals are, and how comparing model behavior with and without context measures real "lift"
• How a lack of lift can signal a skill is now obsolete, saving you tokens and money
• Why a cheaper model with good context can outperform an expensive model without it
• How the Tessl registry adds quality, security, and eval scores to skills
• How Docker sandboxes isolate AI agents using micro VMs, secrets injection, and network rules
• How Tessl "kits" provision tools and skills inside a sandbox for repeatable, shareable setups
🔥 Want More Docker Content?
If you found this demo exciting, hit that like button and subscribe for more! We’ve got even more Docker demos coming your way in this ongoing series showcasing new tools, integrations, and powerful workflows to level up your projects. Stay tuned!
Where to find Docker:
Docker: docker.com
LinkedIn: linkedin.com/company/docker
Bluesky: https://bsky.app/profile/docker.com
X: @docker
Instagram: @dockerinc
