Glossary
Every term, explained once, properly
Plain definitions with the misread, what to do about it, and primary sources. Each entry shows when it was last reviewed.
T4 Orchestration
Agent harness
Everything around the model that turns it into an agent.
T4 OrchestrationWorkflow vs agent
Who decides the next step: your code, or the model.
T3 Tools & protocolsModel Context Protocol (MCP)
The open standard for how AI apps find and call tools and data.
T3 Tools & protocolsMCP vs A2A
Agent-to-tool vs agent-to-agent. Different layers, often used together.
T1 Models & inferenceTime horizon (METR)
How hard a task is at 50% success. Not how long the AI runs.
T5 Evals & reliabilityEvals vs benchmarks
Benchmarks rank models. Evals rank them on your work.
T6 Security & governanceExcessive agency
An agent that can do more than its task needs. OWASP’s #3 for 2026.
T6 Security & governanceLeast agency
The fewest tools, narrowest access and least autonomy that still get the job done.