We document AI replacing human intelligence: how it begins, how it scales, and who gets left behind first.
A mechanism that packages prompts, steps, format constraints, and resources into reusable workflows, marking a shift from temporary prompting to reusable execution units.
It marks the shift from one-off conversation-driven use to reusable workflow-driven execution, allowing experience, formats, and procedures to be stably packaged as execution units.
An AI engineering paradigm centered on building verifiable, iterative, and production-ready execution environments, shifting focus from single-prompt optimization to full-system controllability.
It upgrades the AI development target from "write better prompts" to "build sustainable operating environments," clarifying the core mission of production-grade AI engineering.
A local-first personal AI assistant system that centralizes multi-channel message intake, session routing, and tool execution in a unified gateway runtime.
An AI collaboration framework that turns requirements, specs, design, and task breakdowns into traceable artifacts, making "align on specs first, then implement with AI" an executable process.
An engineering method for structuring task execution around what information to provide to models, when to provide it, and in what volume, marking a shift from prompt optimization to context-system design.
It upgrades the core quality question from "how to ask" to "how to build and manage sustainable context," providing key methodology for complex agent tasks.
A system architecture where multiple software agents collaborate through role specialization to complete complex tasks, entering mainstream production discussion in the LLM engineering era.
It moves AI applications from single-model single-thread execution to parallel multi-role collaboration, significantly increasing the ceiling and scalability of complex tasks.
A broader capability paradigm that extends function calling into multi-tool interaction and orchestration, marking the shift from single API calls to composable tool systems.
It is not a brand-new invention, but a generalized and system-level upgrade of Function Calling that moves AI engineering from single-function invocation to multi-tool collaboration.
An open protocol for connecting models to tools and data through a unified interface, now emerging as an infrastructure-layer standard for AI agents.
It marks the shift from ad hoc API calls to standardized capability access, providing an HTTP-like interface layer for the AI agent ecosystem.
An architecture that combines external retrieval with generation so models can ground outputs in non-parametric knowledge, becoming a key foundation for knowledge-enhanced AI applications.
RAG expanded model memory from "parameters only" to a dual-memory paradigm of "parameters + external retrieval," directly shaping default architectures for enterprise knowledge QA and agent systems.
A capability that lets models call external functions/APIs with structured parameters, marking the shift from text-only generation to executable actions.
An engineering method for stabilizing model outputs through systematic prompt design, marking the shift from ad hoc querying to reproducible interaction programming.