Sequential Agent

Some jobs are too involved for one agent but don't need a free-for-all. Sequential Agent runs several specialists in a fixed order: each one owns a stage, uses its own tools, and passes a finished result to the next. The order is yours, not the model's — so the flow stays inspectable and repeatable, which is what regulated work demands of anything labeled "agent."

How it works

The starter flow is a three-stage research pipeline. Each agent is scoped to one job and equipped with only the tools that job needs:

1. Chat Input states the task — the topic or question to work through.
2. Researcher Agent gathers raw material, using a web search tool to pull current sources.
3. Finance Agent takes the research and does the domain work, using a market-data tool so the figures come from real lookups, not from the model guessing.
4. Editor Agent checks and shapes the analysis into the final, structured answer — with a calculator on hand to verify the numbers before they ship.
5. Chat Output returns it.

The agents pass their output forward in sequence — research feeds analysis, analysis feeds editing. Because the hand-offs are wired explicitly, you can open any stage and see exactly what it received and produced. The three roles are just the example; the pattern is "specialists in a defined order," and you'd re-cast the roles and tools for your own domain.

When to reach for it

Reach for Sequential Agent when a task has clear stages that must happen in order, each benefiting from its own tools and instructions — gather, then analyze, then format; or extract, then validate, then summarize. The fixed sequence trades some autonomy for predictability, which is usually the right trade when the output has to be trustworthy.

When to reach for something else

If the work is a single shaped response with no tools, that's a workflow, not an agent — start from Basic Prompting. If you only need grounding in your documents, retrieval (Vector Store RAG) is simpler and more predictable than an agent. Reach for agents when a step genuinely needs to decide and act with tools — not merely to transform text.

Try it

Run the flow with a task that spans all three stages:

"Research the current state of EU AI regulation, analyze what it means for a
mid-size SaaS company, and write a one-page brief for a non-technical executive."

Watch each agent in turn: the researcher gathers, the analyst reasons over what it found, the editor shapes the brief. Inspecting the hand-offs between stages is the point — you can see where each claim came from.