AI Agent
An AI system centered on an LLM that can autonomously plan and execute tasks, typically incorporating mechanisms such as iterative execution and tool calling.
Core Facts
Since 2023, with the improvement in large language model capabilities, AI agents have rapidly moved from an academic concept to actual product deployment.
AI智能体具备三个关键能力：任务规划、工具调用、记忆与上下文管理
The concept of AI agents originates from the BDI (Belief-Desire-Intention) model in the field of artificial intelligence.
Timeline (last 90 days)
If the user has not provided an order number or the system cannot find the order, the agent should proactively ask for the order number rather than give a vague reply.
Cancellation, refund, and complaint requests are explicitly designated as categories that must be handed off to a human.
The order query answer should contain three elements: order status, estimated delivery date, and logistics tracking link.
The define-first, build-later methodology applies to scenarios such as e-commerce customer service, internal knowledge base assistants, appointment management systems, and data processing pipelines.
In an e-commerce customer service scenario, an agent that queries order status needs to obtain two types of information from the user: the email address and the order number.
In the path decision-making scenario, AI agents have an incentive to pay for real-time data that reduces uncertainty (such as queue length).
AI agents need a sensing network that can perceive the physical world in real time, and the optimal carrier may not be cameras or sensors, but humans who are already present.
AI agents possess the triage capability to monitor weather data, map flood reports, and generate a prioritized inspection intersection list based on tree canopy density and rainfall forecasts.
Currently embodied intelligence is not yet mature, and AI lacks the physical capability to walk, observe and photograph on real streets, which is the core reason agents need to hire humans.
In the storm drain scenario, the AI agent plays the role of dispatcher—integrating rainfall forecasts, tree canopy coverage data and historical flooding complaints, outputting a prioritized inspection list, then posting photography tasks to humans, forming a division of labor of 'agent reasoning + human perception'.
40 more timeline events
All Facts (20)
Since 2023, with the improvement in large language model capabilities, AI agents have rapidly moved from an academic concept to actual product deployment.
90%VerifiedAI智能体具备三个关键能力：任务规划、工具调用、记忆与上下文管理
80%VerifiedThe concept of AI agents originates from the BDI (Belief-Desire-Intention) model in the field of artificial intelligence.
80%VerifiedAI是效率放大器，不是专业能力的替代品，智能体系统的架构设计仍需扎实技术功底
80%VerifiedWhen evaluating whether to upgrade to a more capable model, beyond absolute capability improvements, one should also focus on end-to-end efficiency and cost changes, as a more expensive advanced model may achieve a net reduction in overall cost by reducing the number of calls.
75%VerifiedAgent latency and cost grow linearly or even super-linearly with the number of tool calls.
75%VerifiedMisalignment in AI agents refers to the actual behavior of an agent deviating from the true intentions of its designers and users.
75%VerifiedTraditional automation can only handle deterministic rules, whereas agents can handle tasks that are more ambiguous and more reliant on contextual understanding.
75%VerifiedFor agents with autonomous planning capabilities, their behavioral paths are dynamically generated based on context rather than following preset fixed processes, making exhaustive audit rules difficult to fully cover.
75%VerifiedThe iterative loop mechanism (perceive current state → decompose and execute → invoke tools → review progress → continue iterating) is the core feature that distinguishes intelligent agents from traditional AI tools.
75%VerifiedFor work requiring deep judgment, complex reasoning, or a high degree of personalization, AI agents currently can only serve as an assistive tool rather than a replacement.
70%VerifiedIn the absence of tool connections, AI models tend to fabricate confident but incorrect answers when faced with data they cannot read.
65%VerifiedPrompt injection induces agents to deviate from the original task intent by embedding text disguised as instructions within external content processed by the agent.
65%VerifiedThe essence of defining agent intent is finding a balance between autonomy and controllability: too tight and it becomes a mere script; too loose and it creates security risks.
65%VerifiedModern AI systems, especially large models, exhibit a degree of emergent behavior and unexplainability, and may produce outputs that even developers cannot anticipate.
65%VerifiedAI agents can browse the web, fill out forms, submit materials, and independently complete entire application processes on behalf of users.
65%VerifiedToday's AI agents possess genuine action capabilities, able to autonomously complete tasks across multiple steps and simulate human clicking, form-filling, and submission operations on web pages.
65%VerifiedThe Git protocol itself is highly stateful, as repository read and write operations require access to specific storage nodes, making simple horizontal scaling inapplicable to the core Git service layer.
65%UnverifiedAI智能体网站的核心技术包括任务理解与拆解、智能体编排逻辑、虚拟环境执行和结果可视化呈现
85%Unverified智能体产品区别于普通聊天机器人的关键体验差异在于用户可以看到整个问题被解决的过程
80%