@Yeuoly1: I'm actually curious about what everyone defines as 'an Agent' or 'Agent colleague.' It sounds pretty appealing, but LLMs are fundamentally stateless. What exactly is this entity called 'Agent colleague'? Is it an Agent-Loop with a bunch of m…
Summary
Discusses the definition of AI Agent and 'Agent colleague,' pointing out that LLMs are inherently stateless and questioning the concrete form of the Agent entity.
View Cached Full Text
Cached at: 05/20/26, 06:34 PM
I’m actually quite curious about what exactly people mean by “an Agent” or “Agent colleague” — it sounds pretty sexy, but LLMs are fundamentally stateless underneath.
What exactly is this entity called an “Agent colleague”? Is it an Agent-Loop carrying a bunch of markdown, or is it a session in Claude Code / Codex?
BlanPlan (@blanplan): The idea of not treating Agents as colleagues is correct. The user interface also needs a control surface — instead of mimicking a Slack-style multi-channel team structure, a single conversation entry point to view the agent’s backend progress is more reasonable. Let the agent run large tasks and produce an executable summary, with the user able to intervene at any time.
Similar Articles
@yiliuai: One of my biggest takeaways at MIT is that the world is just a giant makeshift operation. I experienced it deeply again today. This semester I'm taking a course on AI Agents, which requires group projects (the professor said he prefers infra over app layer). Inspired by my previous project connecting an LLM to a vibrator, and because I'm interested in hardware, I originally wanted to build a protocol or middleware layer that allows AI agents to connect and control most hardware. Coincidentally, another MIT undergrad wanted to do something similar; he invited me to join after learning about my vibrator project. But after careful consideration, I abandoned this project and instead chose another more superficial app-layer project. (I don't care about grades anymore, and I couldn't find a project I was both interested in and saw promise for.) My reason for giving up on this direction I was initially very interested in: I believed it had neither commercial value nor any technical moat.
The author shares an experience in an MIT AI Agent course, reflecting on the reasons for abandoning a project aimed at connecting AI agents to hardware, and criticizes a competing project that won high praise from judges and VCs merely by integrating and packaging a simple robotic arm interface.
Beyond LLMs: Why Scalable Enterprise AI Adoption Depends on Agent Logic
IBM Research explores how agent logic—software primitives like knowledge graphs and program analysis—can guide LLM-based agents to efficiently handle complex enterprise workflows, reducing hallucinations and costs while improving outcomes.
@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…
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.
@knoYee_: https://x.com/knoYee_/status/2062780637677752366
The author reviews three months of experience using multi-agent collaboration, summarizing five main pain points (such as conflicts between agents, ignoring boundary conditions, self-censorship failure, difficulty in merging decisions, and exposing harder problems after compressed execution) and two insights (the high value of read-only review agents, and that agent conflicts expose ambiguous requirements), emphasizing the core decision-making role of humans in AI collaboration.
@Potatoloogs: https://x.com/Potatoloogs/status/2057391224592667051
This article deeply analyzes the concept of Agent Harness, which is the engineering infrastructure wrapped around an LLM, including 12 components such as orchestration loops, tool calling, memory systems, context management, etc. The article cites practices from companies like Anthropic, OpenAI, and LangChain, arguing for the critical role of the harness in production-grade AI agents.