Better Work Design Comes First
Multi-agent workflows do not start with more agents. They start with better work design: separating responsibilities so creation, review, judgment, and context each have their place.
Agents Need A World, Not Just A Prompt
Better prompts are not enough for serious AI work. Agents become useful when they can inspect the state of a project, see what changed, and operate inside a shared world.
How To Design Persistent World State For Agents
Agents need more than prompts. This practical guide shows how to give them persistent project state: matters, status, rules, documents, and logs they can inspect and update.
Where Humans Belong In Agent Workflows
AI agents should not be supervised everywhere or trusted nowhere. The useful pattern is human checkpoints: clear places where judgment, risk, and publication decisions stay with you.
How To Run Human Checkpoints With A Software Factory
A practical checkpoint model for agentic work: where to let agents run, where to stop them, and where human approval protects the whole workflow.
AI Writing Needs A Reviewer
Readable AI text is not the same as good writing. The simplest fix is to separate the writer from the reviewer and let human judgment decide what survives.