If you only need a single rewrite, one answer may suffice. When preparing a set of materials, the system may need to read files, identify missing items, call tools, check results, and continue revising—situations that involve how an Agent operates.
Following a Read Result to Proceed
Suppose the lesson-planning Agent reads the course materials first. On success, it can organize a lesson plan accordingly; when it returns "course not found," it should verify the course number or request the correct materials. It should not treat an erroneous result as a successful read.
| Current Result | Reasonable Next Step |
|---|---|
| Materials complete, goals clear | Continue organizing content |
| Missing student prerequisites | Supplement or confirm conditions affecting difficulty |
| Tool execution failed | Investigate cause, try feasible alternatives, or explain limitations |
| Result fits within defined scope | Deliver and state what was checked |
The Model Proposes Actions, the Environment Executes Them
The model can decide the next step is to read materials, the tool executes the actual query, and the application or environment controls whether execution is permitted. The degree of task autonomy and confirmation steps vary by system.
Verify Against Evidence, Not Just Claims
"Already checked" should correspond to actual materials: which files, which documents, which pages. Agents can also make incorrect judgments, so completion criteria and stopping conditions still need to be defined. Don't let unbounded execution continue just because further attempts are possible.
Related Concepts in This Section
- Agent: An AI system that advances tasks toward goals and decides subsequent steps based on process results, typically combining models, tools, and a runtime environment.
For general definitions and technical sources, see the related concepts page. Verification date: 2026-09-09. The examples in this section are for instructional purposes and optional checks for comprehension; they do not restrict continued reading.