Everything Refactix covers across AI and machine learning: generative AI and agents, training and deploying models on your own data, and putting AI to practical use in everyday work. AI is more than LLMs, and the guidance here spans the whole space, focused on what holds up in production rather than the hype.
How Claude MCP tunnels let Managed Agents reach private MCP servers without exposing them publicly. The architecture, the IAM model, and the deployment shape.
Claude Code's /goal command keeps Claude working until a completion condition holds. Agent view puts every background session on one screen. What changed.
Claude Dreaming consolidates agent memory between sessions. How it works, the architecture changes it asks for, and where it actually moves the numbers.
How to package commands, agents, skills, hooks, and MCP servers into a single Claude Code plugin your team or customers install once. Real structure, marketplace.json, and what belongs inside.
How travel companies build rebooking agents that hold up at storm scale. Orchestrator-worker layout, PNR locks, GDS rate limits, and autonomy boundaries that work.
Deep Agents v0.5 async subagents return a task ID instead of blocking the supervisor. The Agent Protocol wiring, mid-task steering, and where the pattern breaks in production.
A production pattern for running parallel Claude Code sessions with --worktree. Syntax, subagent isolation, the shared-state traps worktrees don't solve, and how to clean up.
Running Claude Code in CI/CD without the surprise bills. Headless flags, JSON output parsing, cost controls, and the failure modes nobody writes about.
Claude Code subagents run parallel AI work without context chaos. Config, dispatch patterns, model choice, and the mistakes that waste tokens, with examples.