Browse all articles on AI agent management
Feature Workflow v3 upgrades to a Command + Agent + Skill three-layer architecture. The core insight: template-based workflows are not silver bullets — every project needs purpose-built Skill design.
A summary of the first AgenticOS Workshop at ASPLOS 2026, exploring agents' dual role: as new users demanding fault tolerance, exploratory execution, and resource control; and as new builders enabling specialized software revival and LLM-assisted system management.
Exploring the core of Agent-First transformation: Skill codification, security design, and commercialization methodology, drawing from cross-border e-commerce practices.
A strict linear execution protocol based on STP state anchors. Eliminates hallucination skipping via physical jumps, enabling checkpoint-resumable AI Agent development with absolute idempotency.
AI transformation isn't patching old systems with AI — it's rebuilding organizational DNA. The core is human-AI symbiosis, small-team warfare, value chain compression, platform capabilities, and new management models. Miss any one and you're dead.
AI Coding is not a technical problem — it's a management problem. From requirements to execution to acceptance, building explicit contracts at each stage is the deterministic boundary of AI Coding. A milestone summary from the Agent Management forum.
AgentsZone's roundtable of 19 AI coding pioneers explores how probabilistic LLM systems can ensure code quality through deterministic constraints and independent validation.
CloudMate handles tens of thousands of failure analysis requests per week. This article dissects its evaluate-mutate-backtest loop — how to build a self-evolving Agent system that adapts to unknown failures in a constantly changing production environment.
Lovable, valued at billions, had user debt amounts, home addresses, and API keys extracted in 47 minutes. Software has product managers, architects, developers, testers, and ops — but lacks one role that civil engineering calls a "supervisor."
A response to PingCAP CTO Edward Huang's "Vibe Engineering 2026.1." Huang rewrote TiDB's PostgreSQL compatibility layer with AI, achieving near-production quality code. But he spends 90% of his time evaluating AI output. How do ordinary teams replicate this?
Knowledge-based AIOps systems face a fundamental challenge — software keeps iterating and knowledge expires. CloudMate made a systematic exploration on two fronts: how to keep a knowledge base current with fast-moving code, and how to ensure software evolution doesn't break AI ops effectiveness.
Surveying recent AI operations research to answer three questions: what has AI ops already achieved, where are the current limits, and what comes next. From Microsoft's RCACopilot to NeurIPS's Stratus, a deep dive into Context Engineering and LLM Agent opportunities and challenges.
In 2025, Silicon Valley is swept up in a Forward Deployed Engineer (FDE) hiring frenzy. OpenAI, Anthropic, Databricks and others are all recruiting FDEs. Is this genuinely new, or just a fancy name for the traditional delivery engineer?
Alibaba's qwen-code CLI is a fork of Google's Gemini CLI. Legally fine under Apache 2.0, but from a product management perspective, this was a rash decision. Users discovered the /init command still generates GEMINI.md instead of QWEN.md — a leftover from an incomplete fork audit.
Complaints that AI-written code is unusable? The problem is not correctly applying iterative methodology. Iteration means repeatedly approximating a fixed goal through multiple attempts. In software development, iteration applies to every stage — from requirements to implementation and testing.