Two pieces look at the standing instructions people give AI agents, the information an agent reads before any task begins. Both show people writing down in advance what the agent may do and what it must ask.
Tanner Kohler (20 Sep · 01) of Nielsen Norman Group, a UX research and training firm, studied people who use the AI agent Claude daily for substantial work. Each had built a library of files and connections for Claude, and its information played one of three roles. Global information applies to every task, local information to one project, and ambient information is raw email, transcripts and chat. The global files held templates and brand rules, and governance rules on "Which actions required the user's approval, and which Claude could take on its own". Kohler studied individuals, and advises teams to sort their shared information the same way.
Matt Pocock (20 Sep · 02), an engineer and educator who created the Total TypeScript course, spoke to Gergely Orosz on The Pragmatic Engineer podcast. Pocock writes skills, short instruction files that an AI coding agent reads, and uses them to plan work with agents. His best-known skill, named "grill-me", tells the agent to question the user relentlessly about a plan. He believes agents can now do the tactical programming, which leaves engineers more time for strategic decisions. The account comes from the episode's show notes and describes one engineer's habits.
Both pieces describe people who shape an agent's work before it starts. Kohler's participants keep standing rules, including what needs their approval, and Pocock has the agent question him about each plan. A team that shares such files would also have to decide who may change them, which is an inference from these individual accounts.
Kohler studied individuals using one product, and Pocock describes his own habits. Neither piece shows that written instructions stop an agent from producing wrong work, or that anyone checks the result against them.
A Nielsen Norman Group study found that heavy users of the AI agent Claude sort its background information into three kinds, including standing rules on what needs their approval.
Engineer and educator Matt Pocock uses short instruction files that make AI coding agents question him closely about a plan, which leaves him more time for strategic work.