Tag
The author argues that multi-agent loop failures are caused by poor organizational design rather than prompt engineering, proposing a hierarchical structure with clear authority and termination conditions to prevent indefinite loops.
The author describes a pattern where worker agents emit structured memory events instead of writing directly to shared memory, using a Memory Curator to validate, deduplicate, and route them to appropriate scopes, aiming to prevent memory pollution in multi-agent systems. They compare this approach to existing frameworks and solicit community feedback.
Akshay Pachaar clarifies three distinct AI engineering concepts — prompt engineering (the message), context engineering (the memory), and harness engineering (the machine) — explaining their roles and interplay in building LLM-based agents, with a link to a deeper article on agent harness engineering.
It is recommended to directly view the official Hermes Agent architecture documentation and use Codex or Claude Code to open the project codebase, letting the Agent explain the code for deeper understanding.
Shann Holmberg describes an experimental architecture using gBrain as a shared memory layer for a team of Hermes Agents, allowing specialists to read from a centralized brain before acting and write durable context back.
The author builds two multi-agent AI systems with opposite design philosophies: ChaoticAI (collaborative, org-chart-based) and S.A.G.E. with RAAC (adversarial argumentation). The post shares reflections on memory architecture and the potential synthesis of both approaches.
ActiveGraph introduces a continuity layer for long-running AI agents, building on BabyAGI's concept of persistent state to maintain coherent, evolving models of beliefs, dependencies, and actions over time.
A developer splits their AI agent's LLM calls into a cheap router model (GPT-OSS 120B) for tool-picking and a premium model (gpt-5.4) for synthesis, cutting costs by ~78% while maintaining output quality.
The author argues that the fundamental problem with AI agents lies in LLMs failing to leverage agent environments, requiring separate retraining for each environment and version, which may create release cycle conflicts.
Explores how to architect AI agents for regulated industries like SaMD class II, balancing non-deterministic agent usefulness with deterministic safety zones to satisfy regulatory compliance.
This guide teaches Claude Opus agent architecture to help engineers close the skill gap between $95K and $300K salaries, a skill highly valued by companies.
Describes an agent stack design where a frontier LLM orchestrates fine-tuned small language models for PII redaction, ensuring privacy by keeping raw text local.
The article introduces Needle, a 26M parameter model by Cactus-Compute designed for single-shot tool calling, arguing that tool routing should be separated from reasoning as a structured prediction task to improve agent efficiency and latency.
The author details an experimental multi-agent orchestration framework using a directed acyclic graph (DAG), concentrating intelligence in planner and replanner components while keeping worker agents mechanical. They are seeking community feedback, benchmarks, and existing research to validate its practicality against conventional message-passing approaches.
A tweet highlighting key principles for building agent systems, emphasizing scaffolding, memory, and reusable tools, based on an article by Yohei Nakajima.
This paper introduces Self-Programmed Execution (SPE), an agent architecture where the language model generates its own orchestration program rather than relying on a fixed external harness. It presents 'Spell', a Lisp-based language enabling this self-editing and re-evaluation, demonstrating that frontier models can successfully perform agentic tasks using this method.
A detailed breakdown of Claude Code's six-layer architecture, revealing how it functions as a complex agent harness with input, knowledge, execution, integration, multi-agent, and observability layers beyond just the AI model.
本文深度解析了 Ramp Labs 研发的 AI 代理电子表格工具 RAMP Sheets，涵盖其从流程挖掘起步的演进路径、基于 Excel 公式优先的透明化智能体架构设计，以及在财务自动化场景中的实际应用。