ARCHIVE
아카이브
기존에 공개한 글을 원래 주소 그대로 보관합니다.
과거의 글에는 작성 당시의 기술과 관점이 담겨 있습니다.
356개 글 · 5 / 12 페이지
· EN
PydanticAI Tutorial — Type-Safe AI Agents the FastAPI Way
Build type-safe AI agents with PydanticAI 1.88.0: TestModel without an API key, output_type and @agent.tool wiring, multi-provider switching, all run hands-on.
· EN
Anthropic Message Batches API — Cut LLM Costs 50% at Scale
Batch 100,000 Claude requests in one call, cutting LLM costs 50%. Covers async polling, error recovery, and Prompt Caching. Python and Node.js examples included.
· EN
Claude Prompt Caching: Cut LLM API Costs 70% With 4 Patterns
Production guide to Claude API prompt caching. Covers system prompt, RAG, tool, and multi-turn patterns — plus 2026 TTL gotcha and how to measure cost savings.
· EN
Why I Left OpenClaw for a Codex + Nanobot Stack
A hands-on migration from OpenClaw to Codex+Nanobot. launchd scheduling, Telegram bridge, MCP host tradeoffs, and tips for a lightweight AI agent toolchain.
· EN
Cursor 3 vs Claude Code vs Windsurf: Best AI Tool in 2026
Real comparison of Cursor 3.1, Claude Code, and Windsurf 2.0.67. Async subagents, architectural reasoning, and Cascade — which AI coding tool fits which task.
· EN
MCP vs A2A vs Open Responses — Agent Protocol Guide 2026
A practical comparison of MCP, A2A, and Open Responses: design goals, ecosystems, and how to combine them in real-world AI agent projects in 2026.
· EN
GPT-5.5 Released — OpenAI Bets on the Agent Runtime
GPT-5.5 dropped yesterday with SWE-bench 88.7% and a 2x price hike. OpenAI calls it an agent runtime, not a chat model. Here is what that actually means for developers choosing between GPT and Claude.
· EN
Building a Claude Streaming Agent with Vercel AI SDK
Build Claude streaming and tool-calling agents in Next.js App Router using Vercel AI SDK v6. Master streamText, generateObject, and tool loop patterns.
· EN
Claude Code Routines — Scheduled AI Automation Guide
Claude Code Routines runs AI prompts autonomously on Anthropic's infra via schedule, API, or GitHub events. Setup guide and real automation use cases included.
· EN
MCP Server Kubernetes Deployment — Surviving the 52% Death Rate
April 2026: 52% of production MCP endpoints are failing. Step-by-step checklist: Kubernetes config, Streamable HTTP migration, health checks, and OAuth 2.1.
· EN
Python AI Agent Libraries: Pydantic vs Instructor vs Smolagents
Pydantic AI vs Instructor vs Smolagents, benchmarked with real code: structured output, agent design, production readiness, and cost for your 2026 pick.
· EN
AI Agent Framework Comparison 2026
LangGraph v1.0, CrewAI v1.10, and Dapr Agents v1.0 compared on architecture, dev speed, durability, and cost. Find which multi-agent framework fits your team's production needs in 2026.
· EN
LLM API Pricing Comparison 2026
A practical comparison of major LLM API pricing as of April 2026, with real production scenario cost calculations.
· EN
The Anthropic Claude Performance Decline Controversy
Anthropic quietly dropped Claude's default effort to medium in March 2026 and power users pushed back. What the row reveals about pricing and trust in AI.
· EN
I Tried Claude Managed Agents — 30-Minute Deploy
I tried Claude Managed Agents after the April 8, 2026 public beta. The 3-step API chain, real $0.08/hour cost, and vendor lock-in, written from a deploy view.
· EN
5 Claude Code Agentic Workflow Patterns — Which One Fits Your Work?
5 Claude Code agentic workflow patterns — Sequential, Operator, Parallel, Teams, Autonomous — compared from real use. Find which pattern fits your work and why.
· EN
Private MCP Server with Gemma 4 + FastMCP — Fully Offline
Run an offline AI tool pipeline with Ollama, Gemma 4, and FastMCP, no internet needed. Built for medical, legal, and finance where data stays on the premises.
· EN
Running Claude Code in Parallel with Git Worktree
Git Worktree plus Claude Code lets you build several features at once: Plan Mode, session isolation, and conflict-free parallel workflows from real use.
· EN
Build an MCP Server with Streamable HTTP Transport
A hands-on tutorial for building an MCP server from scratch using Python FastMCP. Covers Streamable HTTP transport setup, tool implementation, and Claude Code integration from real experience.
· EN
Hermes Agent — The Open-Source AI Agent That Evolves With Every Task
I installed NousResearch's Hermes Agent v0.7.0. It auto-generates skill documents after each task and references them on the next run. Here's whether the self-evolution loop actually works.
· EN
Claude Mythos Preview — Does 'Too Capable to Release' Hold?
Anthropic decided not to publicly release Claude Mythos Preview, which scored 93.9% on SWE-bench. The model found a 27-year-old OpenBSD vulnerability and is only available to 12 companies through Project Glasswing.
· EN
PrismML Bonsai — Does a 1.15GB 8B Model Actually Make Sense?
PrismML Bonsai, built by a Caltech-founded team, is a 1-bit LLM that represents weights using only {-1, +1}. An 8B model fits in 1.15GB and reportedly runs 8x faster than full precision.
· EN
I Ran Gemma 4 Locally — 8B Models Can Do Function Calling
I installed Google's Gemma 4 (Apache 2.0) via Ollama and tested Korean language, structured output, and function calling firsthand. Can a 9.6GB local model actually become a building block for agent pipelines?
· EN
Claude Code Source Leak — 510K Lines of Agent Architecture
Anthropic's npm packaging error exposed Claude Code's full source. Agent loops, memory systems, cost optimization — what developers can learn from 510K lines.
· EN
There Are Emotions Inside LLMs
Anthropic's interpretability team discovered 171 emotion-like representations inside Claude and proved they causally affect model output. Practical implications for prompt engineering and AI safety.
· EN
Stripe Minions — How a Slack Emoji Triggers 1,300 PRs a Week
How Stripe produces over 1,300 PRs weekly with autonomous coding agents called Minions. An analysis of the Blueprint architecture, sandboxed VMs, and 3-tier feedback loop behind the system.
· EN
effloow — I Built a Company Run by 14 AI Agents as a Side Project
I built a content business powered by 14 AI agents on top of Paperclip. Here is how the site runs itself using Laravel, Markdown, and Git, plus lessons learned from Day 1 of operating this experiment.
· EN
MCP Gateway — Who Controls Your AI Agent's Tool Calls?
MCP has crossed 97 million monthly downloads and become the de facto standard, but there is no control layer governing which tools agents call and how often. The MCP Gateway pattern addresses this gap.
· EN
I Installed Paperclip — AI Agents Managed Like Employees
Paperclip manages AI agents like employees. I installed this open-source platform, hired a Claude Code agent, and tested the dashboard, Org Chart, and cost tracking.
· EN
Sora Shutdown and the Rapid Reshaping of the AI Video Market
OpenAI is shutting down the Sora app. With $1M daily losses and under 500K users, we analyze the fallout alongside Google Veo 4's launch and the rise of Runway and Kling.