---
description: Three ways of working, explained with an invented packing list for a day trip.
url: https://www.schmalzl.tirol/en/blog/2025-10-13-assistent-workflow-agent
lang: en
date: 2025-10-13
deutsch: /blog/2025-10-13-assistent-workflow-agent.md
---

# Assistant, workflow or agent: who decides the next step?

_13 October 2025 · Archive date_

With AI tools, similar words come up even though they stand for different ways of working. An assistant, a workflow and an agent can all produce text. The decisive question is: who determines the next step?

## A packing list in three variants

An assistant gets a request: "Create a packing list for a one-day trip." It answers, and a person decides whether anything is missing. The collaboration happens in the conversation.

A workflow follows a process fixed in advance. First the length and type of trip are asked, then a suitable template is chosen, and after that the list is created. An AI can also be part of this process; the order stays fixed.

An agent receives a goal and can choose further steps within its permitted options. It could notice that a detail is missing, ask about it and only then revise the list. Which tools it may use for this must be clearly limited. For a text task, access to the necessary content is enough.

This distinction between predefined processes and dynamically chosen steps can also be found in [Anthropic's architecture article](https://www.anthropic.com/engineering/building-effective-agents). In everyday use, however, the terms are not used the same way everywhere.

## More autonomy is an additional decision

For a simple packing list, one good request is often enough. A bigger process brings more points where something can be misunderstood. It becomes interesting when several steps are really needed and their result can be checked.

Before choosing a tool, three questions therefore help: What is the goal? Which decisions stay with the person? How do you know the task is finished? That makes it clear what the agent is actually supposed to do.

## Original finds

These links were the starting point for this post. The commentary in the text is mine; the linked posts are not my own tests.

- [Assistant, workflow or agent? A guide to the terms](<https://www.linkedin.com/posts/felixschlenther_nicht-alles-muss-ein-ki-agent-sein-um-dein-activity-7383383595832332288-Oppr>)
- [Assistant, workflow or agent? A guide to the terms](<https://www.linkedin.com/posts/agenticai-llm-generativeai-share-7449698660830953472-B0h1>)
- [Assistant, workflow or agent? A guide to the terms](<https://www.linkedin.com/posts/felixschlenther_prompt-skill-agent-workflow-was-ist-share-7455503168014020609-1VCn>)
- [Assistant, workflow or agent? A guide to the terms](<https://www.linkedin.com/posts/how-agentic-ai-actually-works-explained-share-7467213368739753984-ZNYE/>)
- [Agents and workflow builders compared](<https://www.linkedin.com/posts/markuskirchmair_kiwoche-kiworkflows-kiagenten-share-7433254094187864065-wVKX>)
- [Five questions to ask about an agent offering](<https://www.linkedin.com/posts/yonathan-cohen_everyones-selling-ai-agents-now-heres-share-7455350809673932800-Jg2L>)
- [Can agents replace individual apps?](<https://www.instagram.com/reel/DX2yCpeTPBW/>)
- [The architecture around a language model](<https://www.linkedin.com/posts/a-harnessed-llm-agent-and-why-llm-is-a-share-7465296752481427456-A8LI/>)

> **About the date:** This post was written on 11 October 2026. The archive date corresponds to the earliest related find in the collection, not to a publication at that time. Technical context and sources added later refer to the time of writing.