
Claude Code Routines: how they work and where the limits are
A routine is a saved Claude Code task that runs in the cloud on a schedule, from an API call or on GitHub events. How to set one up and where the limits are.
What I test and learn about AI coding tools such as Claude Code, AI agents and MCP, and web development with PHP.

A routine is a saved Claude Code task that runs in the cloud on a schedule, from an API call or on GitHub events. How to set one up and where the limits are.

Inception's Mercury 2 doesn't write word by word; it refines a whole answer at once. How that works, how fast it is and where it makes a difference.

Whether an AI agent works reliably depends at least as much on the harness as on the model: context, memory, tools and error recovery. And less is often more.

Mistral Vibe is an open-source AI coding agent from France's Mistral AI. European and open source does not mean your code stays local, though; you have to set that up yourself.

Plugins bundle skills, commands, connectors and subagents. Superpowers makes Claude Code ask questions, plan and write tests first; Anthropic's own plugins do something similar for contracts, bookkeeping and customer support.

OpenCode is an open-source coding agent where you pick the model. What Plan and Build, LSP and AGENTS.md do, which subscriptions you can connect, and why a Claude subscription no longer works since 2026.

You don't need to know every Google AI tool. Gemini with Deep Research and NotebookLM cover most work; Nano Banana, Veo, Opal and the Labs experiments are for when you have a concrete job for them.
In late 2025 NotebookLM gained Gemini 3, tables for Google Sheets, photos as sources and Deep Research. What those features actually do, what's free, and why the tool has been called Gemini Notebook since July 2026.
Claude Code runs as an agent in PhpStorm's terminal. Here is how to set up the CLI and plugin, and the habits that make it work for me: being specific, small steps, reviews and tests Claude runs itself.