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.
Hermes Agent: how Nous Research's self-learning agent really works
Hermes Agent doesn't learn inside the model: after hard tasks it writes its own skill files. How skills, memory and the risks work.
Letting Claude Code work autonomously: auto mode, the Ralph loop and tmux
Letting Claude Code work through the night: auto mode instead of --dangerously-skip-permissions, the Ralph loop against stopping early, and tmux to keep the session alive.
Cognitive debt: when your team no longer understands its AI-written code
Cognitive debt is the gap between how your system works and how well your team understands it. What research says about it, and how to prevent it without ditching AI.
Mercury 2: how a diffusion language model hits 1,000 tokens per second
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.
OpenAI WebSockets: why agents with many tool calls get faster
OpenAI's Responses API has a WebSocket mode. What it does, why agents with many tool calls get up to 40% faster, and when you can skip it.
Programmatic tool calling: why AI agents increasingly call their tools through code
With programmatic tool calling, the model writes code that calls tools, and intermediate results stay out of the context. Sometimes 99% fewer tokens, sometimes more expensive.
Agent harness: why the software around the model decides whether your AI agent works
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.
The 20 best AI sources for 2026, according to my colleague
The list my colleague uses to keep up with AI every morning: twenty sites and newsletters by category, with links checked in September 2026.