Chatbot

Agents (intelligent agents) can invoke tools on a computer and independently complete an entire set of complex tasks, unlike chatbots that can only conduct one-round-at-a-time conversations.

Coding Agent与普通AI聊天机器人的本质区别在于工具调用（Tool Use）能力，包括执行终端命令、读写文件系统、调用API等

There are two main technical approaches in the chatbot field: Rule-based Bots and AI Agents

Ordinary chatbots can only tell users the rough steps, whereas an AI Agent can enter the project directory, create and run scripts, check the output results, and automatically make modifications based on errors.

Chatbot and agent-building platforms are not Agents in and of themselves.

A chatbot is essentially a stateless or short-state request-response system that does not proactively trigger follow-up behavior after each conversation ends.

The brief comments generated by the models were specific to each individual item and remained consistent across multiple runs and across different chatbots.

All chatbot configurations were completely consistent on the sign and ranking of average ratings for administrations.

Cases in which two different configurations gave opposite-sign judgments on the same action within the same day accounted for about 4%.

Differences in answers obtained on different dates were larger than differences within the same day, but the differences grew slowly over time unless the service itself changed.

Most current mainstream chatbots implement multi-turn dialogue by "history concatenation": all past messages are stuffed into the prompt, and the model extracts the relevant information itself.

Competition among chatbots is shifting from "model parameters" to "the refinement of user experience", and differences in multi-turn dialogue, context management and tool calling determine user retention more than basic generation capability.

Current mainstream chatbot products are essentially "instruction-response" systems, in which users bear the structured work of decomposing tasks, organizing context, judging output quality and deciding the next follow-up question.

Ordinary chatbots typically only generate single-turn or multi-turn dialogue and lack the ability to operate on external data and business systems.

The researchers found that newer versions of chatbots showed improvements in answer diversity.

Researchers at the University of Copenhagen compared the answers of 27 chatbots with Google search results across multiple topics and found that the search results contained a noticeably broader range of claims.

Companies can package compliance as innovation: a chatbot that passes a few safety tests does not mean it will not reinforce users' negative emotions or spread misinformation over long-term interaction.

Companies promising that chatbots do not encourage suicide, and platforms restricting children from buying fentanyl, are necessary protective measures, but they are only a minimum safety baseline, not outstanding achievements.

The "check" step is the most substantive dividing line between an agent and a chatbot; a model claiming completion does not constitute evidence, and the result needs to be confirmed correct.

Chatbots output text, whereas WorkBuddy executes operations on the local machine and can connect to external systems, producing actual results rather than suggestions.