Agent state is the live working context an agent carries while doing a task. It may include the goal, what it has already tried, tool results, user preferences, and what comes next. Without state, an agent keeps starting over and loses continuity across steps.

Agent state is the information an agent keeps while it is working. Think of it as the agent’s current situation report. It can include the original goal, the plan, completed steps, failed attempts, retrieved data, selected tools, user preferences, and temporary notes. This matters because most useful work is not solved in one move. If an agent cannot track what happened before, it may repeat actions, forget constraints, or make inconsistent decisions. In product terms, strong state handling makes the experience feel more reliable and less random. A travel agent, for example, should remember that the user wants evening flights, a certain budget, and no long layovers. That remembered context is part of the agent’s state and helps it make better next-step decisions.

The Hidden Variable Behind Smart Agents