AI agents · Types
What is an agent pipeline?
The short answer
The builder fixes the order in advance. Each agent does its part and passes its output to the next. Stages can also run side by side and be combined.
It is one of the 4 forms of multi-agent system, with the manager agent and the agent team and the guardian agent.
| What starts it | Any of the above |
|---|---|
| Who decides the next step | The builder, in advance |
When should you use an agent pipeline?
Use it when the work has known stages that need different skills: extract, draft, check, assemble.
When should you not?
Not when the right order depends on the case.
What does it need to run safely?
It needs a check on the output between stages, a stop when a stage fails, and a record that follows one item through every stage.
An example
A tender response: one agent extracts the requirements, a second drafts answers from past bids, a third checks compliance, a fourth assembles the document.
What else is it called?
Sequential or parallel workflow, prompt chaining.
How do you build one on Katonic?
- New agent, Automate a process. Start blank, fill Identity, and choose a Trigger (Manual to begin with).
- Steps. Add the stages in order: for example a Knowledge step to gather, an LLM step to draft, an LLM step to check.
- Branches & approvals. Add branch to route on a value. Add approval to pause for a person: give it a label and the reviewers.
- Switch to Visual in the header to see the same flow as a canvas. The node list adds an Agent node, "Hand this step to a specialist agent", so a stage can be another agent.
- Guardrails. Turn on the cost ceiling per run and, for write steps, Require human-in-the-loop.
- Test the flow live and read the trace step by step. Then Deploy and Publish.
Questions people ask
What starts an agent pipeline?
Any of the above.
Who decides its next step?
The builder, in advance.
What is a multi-agent system?
A multi-agent system is several agents, each with its own instructions and tools, working on one piece of work. The forms differ in who decides the order. OpenAI advises starting with one agent and moving to several only when needed. Microsoft's guidance is to start with several where the work crosses security boundaries or teams, and otherwise to test a single agent first. Gartner expects 70 percent of AI applications to use multi-agent systems by 2028.