Picking the model is the easiest part of building an AI
agent. The 9 layers around it decide whether it works in production:
- Harness: The loop that lets a model act. Context in, tool call out, result back, repeat
- Memory & state: What it remembers between runs, and where it is in the task right now.
- RAG (retrieval-augmented generation): Pulls the right docs into the prompt so answers come with sources.
- MCP (Model Context Protocol): One standard plug for tools and data. It now lives at the Linux Foundation.
- Skills: Reusable know-how in a SKILL.md file. Only the name and description load until a task needs it.
- Guardrails: Permissions, sandboxes and a human sign-off before anything risky.
- Evals: Scores outputs against what you expected, before your users do it for you.
- A2A (Agent2Agent): How agents from different vendors find each other and hand off work.
- Multi-agent: An orchestrator splits the job. Specialists run in parallel.
MCP, A2A and Agent Skills are all open standards now, so what you build on them
isn't stuck with one vendor. Save this for your next agent build. Which layer is your team skipping right now?