LEAD-01-003-P · Primer — Search vs LLM vs Agent vs Organization of Agents

The idea in one sentence

Search returns information, an LLM returns an answer, an agent returns completed work, and an organization of agents returns a running business process.

Why it exists

Executives conflate these four because they all sit behind a text box, then are disappointed when a chatbot does not run their department. Naming the ladder of outputs sets the right expectation for each tool and shows where the real value lives — in moving from answers to work. It is the difference between a librarian who points you to the right book, a consultant who tells you what to do, an employee who does it, and a functioning department that keeps doing it without you asking.

How it works

Search takes a query and returns a ranked list of documents — you still have to read, synthesize, and decide. An LLM takes a question and returns a synthesized answer — you still have to act on it. An agent takes a goal and returns completed work — it plans, uses tools, executes, and hands you a result to review. An organization of agents takes an ongoing objective and runs a whole process — multiple specialized agents coordinate, hand off, and sustain an outcome over time. The value climbs at each step because the human does less of the doing. Consider hiring: search gives you a list of resumes; an LLM summarizes a candidate; an agent screens applicants, schedules interviews, and drafts scorecards; an organization of agents runs the entire top-of-funnel recruiting pipeline with a human owning final judgment. Critically, evaluation gets harder at each rung — checking a document list is trivial, but confirming a whole process ran correctly is the hardest, most-skipped enterprise problem.

See it

flowchart LR
  A["Search<br/>information"] --> B["LLM<br/>an answer"]
  B --> C["Agent<br/>completed work"]
  C --> D["Org of agents<br/>running process"]

In practice

In fraud operations: search surfaces prior cases with similar patterns; an LLM explains why a transaction looks risky; an agent works a single fraud case end to end and recommends a disposition; an organization of agents runs the entire alert queue, triaging thousands of cases and escalating only the ambiguous ones to human investigators.

The one thing to remember

The prize is not better answers — it is completed work and, ultimately, a process that runs itself under your supervision.