plan-and-execute

Plan-and-execute is a method where, instead of trying to handle a complex request all at once, an AI agent first creates an ordered list of what needs to be done, then executes each item one by one.

It's similar to how a team lead at a company handles a big project: rather than diving in immediately, they first break the overall work into stages and draw up a plan, then assign each stage to a team member, review the results, and move on to the next stage. AI agents work the same way—a part acting as a supervisor draws up the overall plan, while the actual work is split among other parts, each with its own assigned role.

The advantage of this approach is that if circumstances change midway, the agent doesn't have to scrap the entire original plan—it can just adjust the next steps. That's why this approach is especially common in setups where multiple agents divide roles and work together.

In the article, it says the "supervisor agent is implemented as a LangGraph-based state machine and coordinates the overall flow using a plan-and-execute approach." This doesn't refer to the name of a specific AI model, but to the coordination process of scheduling and managing the work order of multiple agents in charge of different tasks.

You can get a feel for the plan-and-execute approach by asking a chatbot something like this:

First break the following task into a step-by-step plan, then carry it out one step at a time and report the results as you go: 'Plan a travel itinerary'

If the response first shows a list of planned steps and then reports the results of executing each step in order, it's mimicking a plan-and-execute structure.