@shao__meng: Three LLM Interaction Paradigms: 1. Web Chatbot 2. Standalone AI App 3. Organization-embedded AI (Claude Tag, Glean Agents) Claude Tag's Core Shift · From 'One AI per Person' to 'One AI per Channel': Team-shared…

X AI KOLs Timeline News
llm interaction-paradigms claude-tag glean ai-agents enterprise-ai

Summary

This article introduces the evolution of three LLM interaction paradigms: from web chatbots to standalone AI apps, and then to organization-embedded AI (such as Claude Tag and Glean Agents). It focuses on Claude Tag's shift from 'one AI per person' to 'one AI per channel', as well as the four pillars of Glean's production-grade independent agents.

Three LLM Interaction Paradigms: 1. Web Chatbot 2. Standalone AI App 3. Organization-embedded AI (Claude Tag, Glean Agents) Claude Tag's Core Shift · From "one AI per person" to "one AI per channel": The team shares a single agent instance with continuous, handoff-capable context. · From "passive response" to "continuous engagement": It remembers discussions, follows up on silent threads, and maintains a long-term presence in the channel. Why Channel-Level Isn't Enough Organizational knowledge is scattered across Jira, Confluence, GitHub, and Slack history. Reading only one channel, the Agent would miss most of the context. The real challenge is building a cross-system, permission-aware, real-time organizational context layer. Four Pillars of Production-Grade Independent Agents (Glean) 1. Identity The Agent has its own identity, permissions, and tool access. Different roles can configure different Agents, with all operations traceable. 2. Memory Learns enterprise runbooks and SOPs, corrects and reinforces from each interaction, accumulating institutional knowledge. 3. Proactivity Does not wait for prompts but actively monitors, flags, follows up, and executes. 4. Accountability Every tool call and decision is visible and explainable, with an "emergency stop" capability to shut down instantly. Practical Example: OnCall Assistant When an alert triggers, the Agent simultaneously reads PagerDuty, Jira, Confluence, GitHub, and Slack, parallel-investigating multiple root causes, drafting fixes, and assigning owners. By the time the engineer opens their computer, the investigation is complete.
Original Article
View Cached Full Text

Cached at: 07/03/26, 12:36 PM

Three LLM Interaction Paradigms:

  1. Web Chatbot
  2. Standalone AI Application
  3. Organization-Embedded AI (Claude Tag, Glean Agents)

Core Changes with Claude Tag · From “one AI per person” to “one AI per channel”: the team shares the same agent instance, with continuous context and handoff capability · From “passive response” to “continuous engagement”: it remembers discussions, follows up on dormant threads, and remains present in the channel long-term

Why Channel-Level Is Not Enough Organizational knowledge is scattered across Jira, Confluence, GitHub, and Slack history. Reading only one channel means the Agent misses most of the context. The real challenge is building a cross-system organizational context layer with permissions and real-time updates.

The Four Pillars of Production-Grade Standalone Agents (Glean)

  1. Identity The Agent has its own identity, permissions, and tool access. Different functions can configure different Agents, and all operations are auditable.
  2. Memory Learns from enterprise runbooks and SOPs, and corrects and reinforces knowledge from each interaction, accumulating institutional knowledge.
  3. Proactivity Does not wait for prompts — actively monitors, flags, follows up, and executes.
  4. Accountability Every tool call and decision is visible, explainable, and includes a “kill switch” capability for immediate shutdown.

Practical Example: OnCall Assistant After an alert triggers, the Agent simultaneously reads PagerDuty, Jira, Confluence, GitHub, and Slack, parallel-investigates multiple root causes, drafts fixes, and assigns owners. By the time the engineer opens their computer, the investigation is already complete.

Similar Articles

This article systematically reviews AI Agent architecture and engineering practices, covering control flow, context engineering, tool design, memory, multi-agent organization, evaluation, tracing, and security. It is based on the OpenClaw implementation and emphasizes the critical role of Harness (testing and validation infrastructure) for system stability.

X AI KOLs

This article systematically reviews AI Agent architecture and engineering practices, covering control flow, context engineering, tool design, memory, multi-agent organization, evaluation, tracing, and security. It is based on the OpenClaw implementation and emphasizes the critical role of Harness (testing and validation infrastructure) for system stability.

@FeitengLi: Built a ReAct agent system by hand: Doing agent systems with LLMs. While walking this evening, I was thinking about how to train an LLM's agentic capabilities, data preparation, model training, constructing RL training with agent trajectory actions, and also about Claude's progress over the past year…

X AI KOLs Following

The author shares their experience building a ReAct agent system and introduces the GLM-5 technical report released by Zhipu AI, which achieves breakthroughs in agentic, reasoning, and coding capabilities.

@cellinlab: https://x.com/cellinlab/status/2064144608242679822

X AI KOLs Timeline

This article introduces the concept of Loop Engineering — instead of directly writing prompts for AI agents, it designs a system (loop) that recursively lets the agent iterate on tasks until completion. The article provides a detailed comparison of how Claude Code and Codex implement five building blocks: automations, worktrees, skills, sub-agents, etc. It suggests this could be the future trend of collaborating with coding agents, but also warns about token costs and AI slop issues.