Derniers articles de la catégorie Agents IA
Kanban, missions, tmux sessions, and metrics: how I built a customizable Agentic OS around Claude Code.
VPS exposure, permissions, scheduled jobs, and prompt injection: a verifiable method for keeping an autonomous agent under control.
How I built a customisable cockpit around Claude Code with Kanban, agents, skills, SQLite and headless/tmux runtimes.
A hands-on test of Claude Code’s /design command, its Artifact workflow, differences from Claude Design, and technical limits.
The same setup and 100% AI-generated code can hide radically different practices. Here is how to separate Vibe Coding from Agentic Engineering.
Your SaaS works for you. That says nothing about what happens with 200 concurrent users. The agentic workflow to load test with k6 and cross-check the result against production errors.
Claude Code killed no-code and agents killed traditional automation. Yet I kept n8n. The real line is no longer no-code versus code: it is stochastic versus deterministic.
VPS, CLIs, skills, kanban: the 4 layers of an AI agent that works 24/7. My crons, my content pipeline, and why CLIs matter more than skills.
The dev job changed: you write decisions, not code. The real risk isn't AI, it's the atrophy of judgment. How to keep the ear that lets you steer.
Your AI agent forgets everything each session. ADRs (Architecture Decision Records) and a context log give it a versioned, queryable memory.
AI makes pentesting trivial. Here are the 3 agents (prevent, fix, monitor) and the hardening rules to lock down your production SaaS.
70× fewer tokens without changing models. An Obsidian vault plus a Graphify graph, shared by every agent I run. The full method.