Glossary

The third-generation AI engineering paradigm. Core formula: Agent = Foundation Model + Harness Layer. Makes AI capabilities controllable, predictable, and scalable through standardized constraints, encapsulation, and orchestration. Includes Dual Harness: Harness for Vibe Coding and Harness for Agent.

The next-generation software paradigm following Software 1.0 (human-written rules) and Software 2.0 (data-trained models). Core characteristic: humans no longer operate software to produce output; AI executes, humans decide. Software evolves from "tool" to "collaborative partner."

A development approach where natural language and structured documents drive AI to generate production-grade code. Its essence is the programmability of engineering culture — encoding team standards, architectural constraints, and quality requirements into AI-executable skill packages via SOP documents.

An AI system capable of autonomously perceiving its environment, making decisions, and executing actions to achieve specific goals. Unlike chatbots, Agents possess autonomy, persistence, tool-use capabilities, and multi-step planning. Enterprise Agents require a Harness layer for governance.

The umbrella term for AI systems with autonomous action capabilities. Distinct from conversational AI (e.g., ChatGPT chat interface), Agentic AI can independently plan task steps, invoke tools, access external systems, and execute complex multi-step workflows. The core technology form of the Software 3.0 era.

An open standard protocol proposed by Anthropic for standardized communication between AI models and external tools/data sources. Analogous to a "USB protocol" for the AI world — enabling different AI models to connect to unified tool and data interfaces.

The middleware between foundation models and business applications in Harness Engineering. Contains: SOP workflow constraints, knowledge base management, Token budget control, model routing, permission management, output validation, and safety boundaries.

Fmode's technical architecture philosophy. The BaaS layer provides out-of-the-box backend services (database, file storage, authentication, push notifications); the PaaS layer provides the runtime environment for AI applications. Together they form the "intelligence hub" — seamlessly connecting AI Agents with enterprise business systems.

Fmode's quantitative model for enterprise AI ROI: Q = (AI Output Value - AI Input Cost) / Traditional Solution Cost. Used to evaluate the economic viability of introducing AI in specific business scenarios. Q > 1 indicates AI is economically superior to traditional approaches.

In GEO and knowledge management, maintaining one authoritative definition page per core concept. AI search engines use these SSOT pages to understand entity relationships between concepts, enabling accurate citation in AI-generated answers.

Content optimization strategy for AI search engines (ChatGPT Search, Perplexity, Google AI Overviews). Unlike traditional SEO (keyword and link-based), GEO emphasizes structured data (JSON-LD), entity relationship definitions, FAQ content coverage, and authority building.

An engineering methodology that defines and controls AI behavior through Standard Operating Procedure documents. SOPs are the core input to the Harness layer — encoding business workflows into structured instructions that AI can understand, execute, and verify.

Fmode's enterprise AI implementation companion model. Day 1: scenario selection and SOP writing; Day 7: first AI execution and results validation; Day 90: routine AI operation for that scenario with replicable experience. Emphasizes rapid start, quick iteration, and pragmatic implementation.

The basic unit of text processing in large language models. ~0.75 English words or ~0.5 Chinese characters per token. AI application operational costs are primarily determined by Token consumption. Token cost management is the core financial consideration for enterprise-scale AI deployment.

Scenario selection methodology (Scenario, Labor, Profit). Evaluates which business scenarios are most suitable for AI introduction: S (Scenario standardization) — how standardized are inputs/outputs; L (Labor) — current human effort required; P (Profit potential) — efficiency gains and cost savings. Higher SLP scores indicate better AI candidates.