Agentic AI Wiki

A concise, source-linked wiki about Agentic AI.

Entities

Concepts

• Agent2Agent Protocol — Open protocol for capability discovery and stateful task coordination among agents.
• Agent Evaluation — Layered evaluation of routing, trajectories, outcomes, safety, efficiency, and adaptation.
• Agent Harnesses — Context, tools, state, controls, evaluation, and operations surrounding an agent model.
• Agent Interoperability — Protocol boundaries for tools, agents, interfaces, and commerce systems.
• Agent Orchestration — Control patterns for fixed workflows, single-agent loops, managers, and decentralized handoffs.
• Agent Safety and Control — Layered guardrails, runtime boundaries, observability, and human escalation.
• Agent Skills — Portable procedural packages with progressive disclosure, evaluation, and lifecycle governance.
• Agent Tools — Data, action, and orchestration interfaces designed for reliable, risk-aware model use.
• Agentic Engineering — Disciplined software development with agent harnesses, specifications, tests, and oversight.
• AI Agents — Definitions, core structure, use-case fit, costs, and evidence status.
• LLM Wiki Pattern — Persistent agent-maintained synthesis built from curated sources, a wiki, and a governing schema.
• Model Context Protocol — Shared client–server boundary for connecting AI applications to data and tools.
• Open Knowledge Format — Portable Markdown and YAML representation for exchanging curated knowledge bundles.
• ReAct — A reasoning–action–observation trajectory format for grounded agent behavior.
• Spec-Driven Development — Versioned specifications as shared contracts for humans, agents, and evaluations.

Comparisons

• LLM Wiki Pattern vs OKF — Maintenance workflow compared with a minimal knowledge-interchange format.
• MCP vs A2A — Tool and context integration compared with stateful agent collaboration.

Queries

Sources

• A Practical Guide to Building Agents — OpenAI’s guide to agent foundations, orchestration, guardrails, and human intervention.
• Agent Skills — Google whitepaper on portable procedural packages, progressive disclosure, and skill evaluation.
• Agent Tools & Interoperability — Google whitepaper on MCP, A2A, A2UI, and commerce protocols.
• Announcing the Agent2Agent Protocol — Google’s draft-era launch rationale and interaction model for A2A.
• Building Effective Agents — Anthropic’s practical taxonomy of workflows, agents, and tool-interface design.
• Introducing the Model Context Protocol — Anthropic’s launch announcement for shared AI-to-data and tool connections.
• Introducing the Open Knowledge Format — Google Cloud’s overview of OKF v0.1 and its relationship to the LLM-wiki pattern.
• LLM Wiki — Andrej Karpathy’s conceptual design for a persistent, compounding, agent-maintained knowledge base.
• ReAct: Synergizing Reasoning and Acting in Language Models — Project-page summary of the ReAct method and reported evaluations.
• Spec-Driven Production Grade Development in the Age of Vibe Coding — Google whitepaper on specifications, production review, and safety controls.
• The New SDLC With Vibe Coding — Google whitepaper distinguishing casual generation from disciplined agentic engineering.
• Vibe Coding Agent Security and Evaluation — Google whitepaper proposing layered security and agent-evaluation frameworks.