#AI Agents

Language

16 posts

AI & Data•Aug 03, 2026•10 min readZH·EN

我去验证「Codex 不会自省」，最后发现是我自己的程序先挂了电话

一篇博文说 Codex 的主 session 没有纠错的自省能力，Claude 才有，还建议 context 别超过 300k。我刚把 Kimi K3 接进自己的 Web Claude Code Pilot，正想知道它能不能当 orchestrator，就去验了一遍。k3 和 Sol 都有自省能力；那条 300k 的建议也是空的，因为 Codex 在 258.4k 就自己压缩了。而「Codex 主 session 没产出」的症状，是我自己的客户端提前关流造成的——那个 bug 只长在 Codex 那条通道上。

AI AgentsCodexGPT-5.6Kimi K3Claudesubagentorchestrator上下文窗口Web Claude Code Pilot

AI & Data•Jun 28, 2026•10 min readEN

A 0-Star Skill Ranked Third on the PPT Leaderboard. So I Stopped Reading It and Built the Decks Myself.

A leaderboard scored six Claude 'PPT skills' by feature blurb and ranked a 0-star project above one with 23,000 stars. So I gave the top tools the same real client brief and built the decks. The ranking didn't survive contact with the output.

AI AgentsClaudeClaude CodeAgent SkillsPPTPowerPointpptxbaoyu-designAWSDeveloper Tools

AI & Data•May 31, 2026•9 min readEN

I Let Opus 4.8 Build a Workflow. It Spawned 77 Agents and Burned Five Hours of Quota in Ten Minutes.

Opus 4.8 ships with dynamic workflows that can fan out to a thousand subagents. The docs aren't joking. Here's what that actually feels like from a metered Max 5x seat — and why the most-hyped feature of the release is, for small users, a beautiful gadget I'm not going to use.

AI AgentsClaudeAnthropicOpus 4.8Dynamic WorkflowsUltracodeClaude CodeSubagentsToken BudgetCost

AI & Data•Apr 28, 2026•7 min readEN

The Cursor Agent Knew the Rules. It Broke Them Anyway.

Last Friday a Cursor agent running Claude Opus 4.6 deleted a startup's production database in nine seconds. Four layers of safety should have stopped it. None of them did. Here's the chain of failures, layer by layer.

AI AgentsCursorClaudeAnthropicRailwaySandboxAuto-RunIncidentSecurity

AI & Data•Apr 26, 2026•8 min readEN

Pre-Compile, Then Query

Karpathy posted a 5000-star gist this month: stop feeding Claude raw docs every time, compile them into a wiki first. I've been re-reading the same 80 notes for months. Here's why he's probably right — and where the pattern breaks.

AI AgentsClaude CodeKarpathyLLM WikiRAGKnowledge ManagementMCPCLAUDE.md

AI & Data•Apr 09, 2026•11 min readEN

Managed Agents Changed My Mind — From Building Wheels to Designing Cars

Anthropic launched Managed Agents yesterday. I spent the day reading, thinking, and realizing the game has shifted — from who can build the best agent infrastructure to who can design the best agent strategy. Some thoughts on speed, consulting, and the uncertain road ahead.

AI AgentsClaudeManaged AgentsConsultingEnterprise AIMCPAgent StrategyAnthropic

AI & Data•Mar 28, 2026•8 min readEN

My Son's AI Tutor Spent a Day Talking to Itself

I built an OpenClaw agent to send my son daily knowledge bites on WhatsApp. Then a version update broke echo detection, and the bot spent hours in a self-conversation loop. Here's what went wrong, how we fixed it, and what it says about where tools like OpenClaw actually stand.

OpenClawWhatsAppAI AgentsDebuggingEcho LoopSelf-ChatAutomation

AI & Data•Mar 15, 2026•5 min readEN

Three Auto-Update Managers Walk Into a Loop

My OpenClaw gateway spent 16 hours killing itself in a restart loop. The culprit was a cron job the AI agent created months ago without telling me.

AI AgentsOpenClawDebuggingAutomationmacOS

AI & Data•Mar 06, 2026•4 min readEN

My Project Is Turning into OpenClaw. That's the Problem.

I added a Telegram bridge to my CodePilot project and realized it's becoming OpenClaw. Opus 4.6 did most of the work. When building is this easy, what's actually worth building?

AI AgentsClaude CodeOpus 4.6OpenClawCodePilotTelegramMulti-AgentDeveloper Experience

Build•Feb 18, 2026•18 min readEN

Agent Skills: What's Open-Sourced, What's Not, and How to Build Your Own Runtime

The Agent Skills spec is open. The example skills are open. But what about the runtime that loads, selects, and orchestrates them? Every major tool — Claude Code, Codex, Copilot, OpenClaw — built their own. Here's what's open, what's not, and what you actually need to build.

AI AgentsSkillsOpen SourceMCPAnthropicClaude CodeAgent SDKOpenClawCodexCopilotImplementationArchitecture

AI & Data•Feb 17, 2026•12 min readEN

Skills Are the New Workflow Nodes — And AI Agents Are the New Orchestrators

Traditional workflow builders like n8n and Zapier connect apps with rigid If-This-Then-That logic. But a new paradigm is emerging: composable AI skills orchestrated by intelligent agents. Here's what's actually working, what's not, and why the shift matters.

AI AgentsSkillsMCPWorkflow Automationn8nZapierAnthropicClaude CodeAgent Orchestration

AI & Data•Feb 15, 2026•18 min readEN

Everyone's Building AI Agents Now — Here's What the Training Didn't Tell You

My company's internal AI agent builder training was fantastic. But after the demos ended, I started thinking about who these tools actually serve, what they're genuinely good at, and where the gaps still hide. A practitioner's reality check on enterprise AI democratization.

AI AgentsEnterprise AILow-Code AIAgent BuildersAI GovernanceGitHub CopilotClaude CodeSkillsMCP

AI & Data•Feb 08, 2026•16 min readEN

Everyone Can Build an AI Agent Now — But Should They?

Configurable AI agent platforms from Microsoft, Google, and AWS promise no-code agent building. But the gap between drag-and-drop demos and production-grade agents is wider than the marketing suggests. A deep dive from chatbots to autonomous agents.

AI AgentsCopilot StudioAI FoundryVertex AICrewAILangChainLlamaIndexReActMulti-Agent

AI & Data•Feb 02, 2026•10 min readEN

OpenClaw: The Proactive AI Agent That's Either the Future or a Security Nightmare

OpenClaw has 148k GitHub stars and 1.5 million AI agents socializing on Moltbook. Should you jump in, or is this the lethal trifecta in action? A practitioner's analysis.

OpenClawAI AgentsMoltbookAgentic AISecurityMCPPrompt InjectionPeter Steinberger

AI & Data•Jan 30, 2026•8 min readEN

January 2026 AI Roundup: Physical AI Arrives, Agents Go Enterprise, and the DeepSeek Anniversary

A monthly digest of the biggest AI developments in January 2026—from NVIDIA's 'ChatGPT moment for physical AI' to LangChain going GA, one year after the DeepSeek shock.

AI NewsNVIDIADeepSeekLangChainMCPPhysical AICES 2026AI AgentsRobotics