
The author split article work into planning, structuring, drafting and fixing sub-agents in Claude Code, plus a still-separate fact-checking role, and put three review checkpoints — after planning, after structuring, and before publishing — where only the author decides pass or rework.
Summaries like this, in your inbox every morning.
The setup rests on a deliberate separation of roles: writing, fixing and doubting (fact-checking) are never merged, because the author believes mixed roles lower accuracy. The doubting role is the exception — it has not been turned into a sub-agent and is instead handled by the AI in the ordinary conversation where the author talks to Claude Code directly.
Deciding who passes and who is sent back is the author's alone, and the idea of human stop points predates having the AI write the system's own specification.
One practical trap surfaced only after the build. The author did not notice. Definitions load from the launch directory and from directories added with --add-dir, so the first thing to check when a not found error appears is where the session was started.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
Circlia started a proof-of-concept for TASTE SIGNAL: visitors to Holland Village Cafe FUKUOKA taste cookies an…

In a webinar hosted by French Mathematical Society president Isabelle Gallagher, Fields Medal winner Hugo Dumi…

NVIDIA released Nemotron 3.5 Super VL, an open-weight model with 120 billion parameters that handles text, ima…

NVIDIA released Nemotron 3.5 Bolt 74B-A9B 24GB, an AI model for local agents that hits over 80 tokens per seco…

NVIDIA published a blog post titled "AI Security Is an Engineering Problem - How to Solve It at Every Layer of…

In the test, Haiku 5.5 on medium setting averaged scores close to Sonnet 5.5 on routine documents at about 1/1…
