YTYK You think you know

Card 18 of 18 Agents ●●○ Working knowledge

Most tasks are not agent tasks

If the steps are known in advance, a pipeline does the job more cheaply, faster and more testably than an agent. Agency earns its cost when the path genuinely cannot be specified up front — open-ended investigation, variable inputs, dead ends that need backing out of.

Why it is like this

An agent buys adaptability by giving the model the control flow. If you already know the control flow, you are paying for a decision you had already made, and giving up determinism to do it.

Not to be confused with

Old-fashioned. Most shipped GenAI value is single calls and fixed workflows; agents are the smallest and most expensive tier, and the one most often chosen for its name.

In practice

The test: can you draw the steps? If yes, write them. If the drawing has a box labelled "figure out what to do next", that box is the agent, and it may be the only part that needs to be.

Source: OWASP Top 10 for Large Language Model Applications, 2025 — LLM06:2025 Excessive Agency. Read on .

The OWASP Top 10 for LLM Applications is © the OWASP Foundation and published under a Creative Commons licence; NIST AI 600-1 is a US Government work in the public domain. These cards are independent commentary written in our own words. ytyk.org is not affiliated with or endorsed by the OWASP Foundation or NIST.

Odd one out Check yourself

Three of these justify an agent over a fixed pipeline. Which does not?

  1. The path to the answer varies with what is found along the way
  2. The task requires backing out of dead ends
  3. The inputs are open-ended and cannot be enumerated
  4. The task has many steps
Show the answer

The task has many steps

Many steps is an argument for a pipeline — they are cheaper, faster and testable. Unpredictable steps are the argument for an agent.