ARCHIVE
아카이브
기존에 공개한 글을 원래 주소 그대로 보관합니다.
과거의 글에는 작성 당시의 기술과 관점이 담겨 있습니다.
356개 글 · 6 / 12 페이지
· EN
LiteLLM Supply Chain Attack — AI Dependency Blind Spots
Deep analysis of the LiteLLM supply chain attack on PyPI: dependency vulnerabilities, malicious package detection, and a defense checklist for AI engineering teams.
· EN
AI Coding Agents Leaked 29 Million Secrets
According to the GitGuardian 2026 report, repositories using AI coding tools leak secrets at twice the GitHub average. Over 24,000 credentials were exposed in MCP config files alone.
· EN
Mistral Voxtral TTS — 3-Second Voice Cloning, Open Weight
Analyzing Mistral's 4B open-weight TTS model Voxtral. It beat ElevenLabs in human evaluations but lacks Japanese support, a dealbreaker for Asian markets.
· EN
Building Real-Time Voice Agents with Gemini 3.1 Flash Live
Analyzing Google's Gemini 3.1 Flash Live for building real-time voice and vision agents. Covers API structure, tool calling, 90+ language support, and honest limitations from a developer's perspective.
· EN
GitHub Copilot Will Train AI on Your Code Starting April
GitHub announced that Copilot Free/Pro/Pro+ interaction data will be used for AI model training by default. Here is how to opt out and what it actually means.
· EN
Google TurboQuant: 3-Bit KV Cache With Zero Accuracy Loss
A deep dive into Google TurboQuant's PolarQuant and QJL techniques — 6x KV cache memory reduction and 8x attention speedup, and what that actually means in practice.
· EN
Vibe Physics — A Physics Professor Had Claude Write a Paper
Analyzing Anthropic's Science blog debut where Harvard physicist Matthew Schwartz supervised Claude as an 'AI grad student.' 110 drafts, 36M tokens, and a paper completed in two weeks.
· EN
Dapr Agents v1.0 GA — How to Make AI Agents Survive in Kubernetes
Analyzing Dapr Agents v1.0 announced at KubeCon Europe 2026 — its durable workflows, automatic recovery, and scale-to-zero — and how it differs from existing agent frameworks.
· EN
NemoClaw — NVIDIA Wraps OpenClaw with Enterprise Security
Announced at GTC 2026, NVIDIA NemoClaw is an open-source reference stack for running OpenClaw safely in enterprise environments. A look at its realistic limitations and possibilities in alpha stage.
· EN
Coding via Telegram with Claude Code Channels
Claude Code now has a Channels feature. Send a message on Telegram, and Claude running in your local terminal executes code and replies.
· EN
Deeptune: 'Training Gyms' for AI Agents
Deeptune raised a $43M Series A led by a16z. Their approach — training AI agents through RL environments that simulate professional workflows — signals a fundamental shift for engineering organizations.
· EN
IBM's $11B Confluent Buy — Real-Time Data Fuels AI Agents
IBM acquired Confluent for $11 billion, elevating real-time data streaming as core infrastructure for AI agents. A CTO-level analysis of what this deal means and how engineering orgs should respond.
· EN
Gemini Embedding 2 — How Multimodal Embeddings Change RAG
Google's first native multimodal embedding model: what shifts from text-only embeddings, how retrieval changes, and how to use it in a production RAG pipeline.
· EN
AlphaEvolve Breaks 5 Ramsey Records: AI as Research Partner
Google DeepMind's AlphaEvolve broke five Ramsey number records held up to 20 years. What this means for AI as a research partner and engineering leadership.
· EN
Hindsight — Open-Source MCP Memory That Gives AI Agents Learning
Analyzing the architecture, core capabilities, and production deployment strategies of the Hindsight MCP memory system that solves the AI agent memory problem.
· EN
Morgan Stanley's 2026 AI Leap Warning — 5 Things CTOs Must Do Now
Morgan Stanley predicts a non-linear AI capability leap in early 2026. Here are 5 strategies engineering leaders should execute right now to stay ahead.
· EN
Perplexity Computer — The Rise of Always-On AI Agents
Analyzing Perplexity Personal Computer and Enterprise. A 24/7 always-on AI agent completed 3.25 years of work in 4 weeks — with EM adoption strategies.
· EN
AI Agent Identity Dark Matter: Why Enterprises Lose Control
70% of enterprises run AI agents in production, yet 80% cannot see what they do. What identity dark matter is, why risk compounds, and 5 controls to apply now.
· EN
GLM-5: MIT Open-Source Frontier Model for Enterprise
Zhipu AI released GLM-5 with 744B MoE trained on Huawei Ascend without NVIDIA. A cost-effective MIT-licensed alternative for enterprise deployments.
· EN
Microsoft Agent Framework GA: AutoGen + Semantic Kernel Unified
Microsoft Agent Framework, unifying AutoGen and Semantic Kernel, is approaching Q1 2026 GA. From an EM/CTO perspective, this post covers key features, migration strategy, and a production adoption roadmap.
· EN
OpenAI Acquires Promptfoo — The AI Agent DevSecOps Era Begins
OpenAI acquired Promptfoo (25% of Fortune 500). Learn what it means for AI agent DevSecOps: red team testing, CI/CD pipelines, and behavior policies.
· EN
How to Detect Agent Washing: A 7-Point EM Checklist
~130 of thousands of AI agent vendors build truly agentic systems. Use this EM 7-point checklist to test goal re-routing, context memory, and tool flexibility.
· EN
Anthropic's Big AI Ecosystem Play — Institute & $100M Fund
Anthropic Institute launch, $100M Claude Partner Network, and Claude Certified Architect. A CTO-level analysis of AI vendor ecosystem maturity in 2026.
· EN
GPT-5.4 — Native Computer Use and the 1M Context Window
OpenAI released GPT-5.4 on March 5, 2026. Computer use surpassing humans (75% vs 72.4% on OSWorld), 1M token context window, 47% token savings via tool search — here's what engineering managers need to know.
· EN
9 Design Principles for Production-Grade AI Agent Deployment
Solve the core challenges of deploying AI agents to production in 2026 with 9 battle-tested design principles from arXiv research, presented from an Engineering Manager perspective.
· EN
AI Agent Observability in Production
A practical guide for Engineering Managers on monitoring multi-agent LLM systems in production. Covers distributed tracing, metrics, logging, OpenTelemetry, and a comparison of Langfuse, LangSmith, and Braintrust.
· EN
MCP Apps: Interactive UI Running Inside AI Chat
How MCP Apps transform AI agent UX—from sandboxed iframe and JSON-RPC bidirectional communication architecture to real implementation code. A complete guide from an Engineering Manager perspective.
· EN
mcp2cli — Cut MCP Token Costs by 96–99% with CLI-Based Tool Discovery
Connecting MCP servers injects all tool schemas into context every turn—362,000 tokens wasted for 120 tools over 25 turns. mcp2cli solves this with CLI-based on-demand discovery, cutting costs by 96–99%.
· EN
OpenAI Open Responses: A Common Standard for Agentic AI
OpenAI Open Responses spec standardizes agentic AI workflows. We analyze its core concepts, ecosystem support, and adoption strategies from an EM/CTO perspective.
· EN
Claude Code Review — Multi-Agent PRs Lift Coverage to 54%
Complete breakdown of Anthropic's new Code Review feature for Claude Code: parallel multi-agent architecture, $15–25 per-PR cost structure, and everything Engineering Managers need to know before adopting