Run loops in agent-native design <p>In this channel's material on building <b>agent-native products</b>, the term <i>run loop</i> comes up as part of the vocabulary Mike Krieger uses when describing how software built around AI agents behaves differently from traditional applications — where a product responds to a user action, does one thing, and stops.</p><p>In that framing, an agent-native product keeps <b>running in a loop</b>: it takes in context, acts on it, and continues working on the user's behalf rather than requiring a new instruction for every step. This is one of the design shifts Krieger discusses when explaining what changes when you build for agents — and what questions, like personal agents, will define the near future.</p><KeyPoint>- A <b>run loop</b> describes the ongoing cycle of an agent working on a user's behalf, rather than a one-shot request-and-response interaction.</KeyPoint>- It appears in the interview as part of the broader contrast between traditional products and <b>agent-native</b> design.</KeyPoint>- Understanding it helps when evaluating how agentic browsers and agent-native products actually behave.</KeyPoint>