Practical agentic AI, weekly. System prompts, skill reviews, and configs you can actually use.
Jul 10, 2026
•
6 min read
The 2026-07-28 spec deletes the session, so any state your tools keep has to move somewhere the model can see it
Jul 3, 2026
An approval gate is the one checkpoint that stops an autonomous agent before it does something you can't undo
Jun 26, 2026
7 min read
Outcomes gives a Managed Agent a separate grader that rereads the work against a rubric you write, and sends it back until it passes
Jun 19, 2026
Claude Managed Agents run on Anthropic's infrastructure, on a schedule, while your machine is asleep
Jun 12, 2026
MCP Apps let a single tool return an interactive UI that renders right in the chat
Jun 5, 2026
4 min read
When to spin up an agent team, and when one agent is still the right call
May 29, 2026
5 min read
The same SKILL.md runs in 20+ agents. Here's how to write one that travels.
May 22, 2026
One context window is a ceiling. Here's how to factor the work across a crew and cut your token bill ~40% doing it.
May 15, 2026
Skills are advisory. Hooks are statutory. Three production patterns that take you from 20% to 84% rule enforcement.
May 8, 2026
The description engineering problem nobody mentioned.
May 1, 2026
Why long Claude Code sessions degrade at hour three — and the subagent pattern that fixes it.
Apr 24, 2026
The 5-layer threat model, a 5-minute audit, and the two rules that stop 90% of attacks.
Apr 17, 2026
The 4-layer memory stack that fixes it
Apr 10, 2026
9 min read
Stop writing prompts. Start being interviewed.
Apr 3, 2026
8 min read
A 5-section framework, a copy-paste template, and the 3 mistakes everyone makes
