Stop calling everything an “AI agent.”

An agent is not a tool.
A tool is not a skill.
A skill is not a harness.
A plan is not a spec.
A critic is not a verifier.

These are different layers with different jobs.

The cheat sheet breaks down the vocabulary behind agentic software: agents, harnesses, skills, specs, plans, adversarial loops, MCP, A2A, context engineering, evals, and the controls that make actions reliable.

The three questions worth asking in every architecture review:
Who decides?
What can it do?
What proves it worked?

Save this for your next design discussion.

Which pair of terms gets mixed up most on your team?

#AIEngineering #SoftwareEngineering #AIAgents
