In this section, focus is made on what can be done to reduce vendor dependency and easily switch between LLM providers.

Use standards

Use standards to reduce vendor dependency and ensure easier switches to competitors or self-hosted alternatives without losing time on team re-setup.

• Agent Client Protocol: Any LLM on any editor (or extension), switch LLM providers without changing your setup.
• Agent Skills: Build standardized skills for all agent runtimes to use.
• AGENTS.md: A simple instruction filename for all agent runtimes (well almost).
• Model Context Protocol: Connect any agent runtime to any compatible external system.
• Agent Plugins: Share components in one format for all agent runtimes to use.

Favor Open Source

Favor Open Source projects, components, agent runtimes and tools with no LLM provider lock-in.

• agentgateway: Open source proxy for MCP, Agent-to-Agent, LLM and API traffic, providing drop-in security, observability and governance.
• Goose: Open source, extensible AI agent runtime (CLI, desktop app, API) that works with any LLM provider.

Both projects live under the Agentic AI Foundation (AAIF), a newly founded (December 2025) neutral foundation, Linux Foundation-backed, for open agentic AI standards and projects.

What’s starting here is very similar to the Cloud Native Computing Foundation (CNCF).