Agent
An Agent, or intelligent agent, is a program or system in artificial intelligence that can perceive its environment, make autonomous decisions, and take actions to accomplish goals. Its core characteristics include autonomy—operating without continuous human intervention—goal-directedness, enabling the planning and execution of multi-step tasks, and tool-calling capability, allowing interaction with external systems or APIs. Agents are widely used in task automation, dialogue systems, and programming assistance.
Core Facts
Mainstream Agent architectures typically use a large language model as the brain, combined with memory modules, planning modules, and tool interfaces to accomplish complex tasks
Agents possess capabilities for planning, tool invocation, memory, and autonomous iteration, distinguishing them from the single-round question-and-answer mode of traditional large language models.
A Skill is a modular encapsulation that packages specialized capabilities, domain knowledge, and operational norms into reusable units, enabling Agents to invoke them automatically at the appropriate moment.
Timeline (last 90 days)
In the agent loop, a large share of token consumption comes from the agent repeatedly making small judgments (such as safety gates, format checks, and whether to hand off to a human), rather than from executing the task itself.
Under the current architecture, each small judgment means a complete generation process, including input prompt, output reasoning, text parsing and format validation.
When operating in an Agent window, it is best to enter the corresponding working directory, because Agents have a workspace concept and matching the directory makes operations more accurate.
AIGC provides suggestions, plans or content that require subsequent manual processing; agents deliver executable results, such as code generation, testing and deployment execution, or sending emails, compiling reports and querying databases.
AIGC responds passively and depends on precise prompts; agents plan proactively and make decisions autonomously, trying to give the optimal solution through workflows.
In agent scenarios, the focus of prompt engineering shifts from "polishing the wording of a sentence" to "designing tool sets, planning flows and setting constraints", which is the deep reason why general-purpose tools are hard to reuse directly and must be encapsulated a second time.
AIGC mainly interacts through text (and images); agents can interact with the real world through tools, such as ordering takeout or buying high-speed rail tickets.
If an agent designer cannot answer "which anchor will the agent lose first," they only understand the window size and not the agent's actual behavior.
After migration, you must not blindly let the Agent continue execution; instead, you should first compare the code changes recorded in the cloud session with the actual state of the local workspace, and manually sync branches or install missing dependencies if necessary.
Developers should first review the code changes after migration, rather than directly letting the Agent continue execution.
40 more timeline events
All Facts (20)
Mainstream Agent architectures typically use a large language model as the brain, combined with memory modules, planning modules, and tool interfaces to accomplish complex tasks
90%VerifiedAgents possess capabilities for planning, tool invocation, memory, and autonomous iteration, distinguishing them from the single-round question-and-answer mode of traditional large language models.
90%VerifiedA Skill is a modular encapsulation that packages specialized capabilities, domain knowledge, and operational norms into reusable units, enabling Agents to invoke them automatically at the appropriate moment.
90%VerifiedA complex Agent task may require dozens or even hundreds of model calls
90%VerifiedSince 2023, with the leap in reasoning capabilities of large language models such as GPT-4 and Claude, Agents have rapidly moved from academic concepts to engineering practice.
90%VerifiedLangGraph abstracts agent workflows as directed graphs, where each node represents an agent or processing step, and edges represent state transition conditions.
90%VerifiedAgent planning capabilities are typically implemented through the ReAct (Reasoning + Acting) framework.
90%VerifiedA typical Agent is composed of a cycle of 'Perception—Planning—Action—Reflection'.
85%VerifiedAI evaluation targets are divided into three layers—large language model evaluation, knowledge base (RAG) evaluation, and agent (Agent) evaluation—forming a progressive relationship of increasing complexity and knowledge depth.
80%VerifiedAn agent's long-term memory can be subdivided into three categories: episodic memory, semantic memory, and procedural memory.
80%VerifiedDeepSeek Harness adopts a plugin-based architecture in which the model, tools, session sandbox, and the Agent main loop itself can all be swapped out and recombined.
80%VerifiedCode-execution-type agents have a complete "perception–planning–execution–feedback" loop capability, enabling them to read data, autonomously formulate experimental plans, execute code, and dynamically adjust strategy based on results.
80%VerifiedThe role of testers is shifting from executors to AI reviewers, and the prerequisite for harnessing AI is possessing foundational skills such as automated testing and performance testing.
80%Verified如果一个10步操作中每步准确率为95%，整体成功率仅约60%
80%VerifiedMCP（Model Context Protocol）是Anthropic提出的开放协议，旨在标准化大模型与外部工具、数据源之间的交互方式
80%VerifiedLangGraph is an Agent orchestration framework launched by the LangChain team; its core concept is to model the Agent's execution process as a directed acyclic graph (DAG), supporting complex control flows such as loops and conditional branches.
80%Verified一个典型的Agent架构包含大模型核心、工具集、记忆系统和规划模块四个核心组件
80%VerifiedAn Agent's token consumption for a single task may be 5–10 times that of an ordinary conversation, due to the involvement of multiple rounds of large model calls.
80%VerifiedAI Agent基于ReAct（Reasoning + Acting）等框架，能够将复杂任务分解为多个子步骤，自主决定何时调用外部工具
80%VerifiedAI Agent具备'规划-执行-反馈'的完整闭环能力，能自主拆解复杂任务、调用外部工具、持续迭代直到任务完成
80%