Tag
The author argues that agent tool 'success' responses only prove a request was accepted, not that the intended state landed, and recommends verifying every consequential write with a fresh read from the system of record through a different path, checking the specific fields that matter.
A long-form engineering article discussing harness design for vertical-domain Agents: how to integrate project knowledge, workspaces, identity authentication, and long-task recovery on top of a Deep Agents foundation, and how to iterate on adding or removing capabilities based on evaluation results.
作者跟进此前关于不信任工具调用返回200的讨论，提出AI代理说"done"时实际可能对应六种状态（意图、尝试、接受、应用、正确、生效），并建议对涉及发送/支付/删除的操作进行系统对计划、结果对原始请求的双重校验，避免把"已接受API"误报为"已完成"。
Matthias Endler-style (matklad) blog post arguing micro benchmarks should be tuned to run around 300ms — fast enough to iterate quickly, slow enough to avoid startup overhead and to give humans an intuitive sense of speed — with the aim that benchmarking provides decision-making intuition rather than precise measurement.
Sophie Alpert argues that there are no lossless transformations of natural-language text, so engineers should stand behind every idea and sentence when using AI to help write docs, and not disclaim AI-generated content.
Anthropic's engineering team shares their complete experience from over two years of security isolation for Claude products, including three isolation modes (ephemeral containers / HITL sandbox / local VM) and six incident postmortems with fixes, emphasizing environment-layer defense over model-layer.
A document titled 'Deep Learning Foundations, Architectures & Engineering Practice' is shared, likely covering fundamental concepts, architectures, and practical engineering aspects of deep learning.
Harness Engineering is not mysticism, but an engineerable living product. The article proposes a six-layer engineering framework (Rule, Skill, Sub Agent, Workflow, Scripts, dev-map), emphasizing starting simple, relying on scripts rather than prompts, and improving through iteration.
An open-source 65-line Markdown file that summarizes 4 programming rules for Claude to improve code quality and save tokens. It has garnered 130k+ stars on GitHub, inspired by Karpathy.
The user asks about the best agent memory solution currently in engineering practice and mentions Hermes Agent.
Introducing hello-agents, an open-source project that topped GitHub trending. It systematically organizes courses on AI and Agents from theory to practice, covering skills like Agentic RL, SFT, GRPO. Created by former Vercel engineer Mat, it distills engineering experience into 16 skills, such as Grim prompting techniques and red-green cycle testing methods.